Artificial Intelligence Integration & Impact

8/7/2026

Introduced Radar Researcher, an AI-powered tool that allows users to query Cloudflare Radar's extensive Internet data using natural language. This tool leverages Cloudflare Workers, Durable Objects, SQLite, Workers AI (with model fallback), AI Gateway, and the Cloudflare MCP server with Code Mode to translate natural language queries into API calls, retrieve data, and render interactive charts. It also supports vision-capable models for analyzing chart screenshots and provides an auditable trace of its reasoning process.

8/7/2026

This post details the convergence of Cloudflare's Workers AI and AI Gateway products into a single AI control plane. Key technical contributions include: unifying the Workers binding and REST API entrypoints to a single `/ai/` endpoint, enabling automatic creation of a default AI Gateway for Workers AI users to provide observability and logging without explicit setup, introducing unified billing for AI Gateway credits to be used across all supported model providers including Workers AI, and outlining future capabilities like model-first routing (requesting a model by capability rather than provider) and intelligent routing (gateway automatically selecting the best model for a task).

8/6/2026

This post introduces Cloudflare AI Search, a new product that simplifies the integration of search capabilities into AI agents and applications. It automates the process of indexing data (files, websites), provides unified search endpoints with custom domain support, and integrates with EmDash CMS. The post details how AI Search powers internal tools like the Cloudflare Dev Stack MCP, which uses multiple AI Search instances for comprehensive documentation search. It also previews pricing for AI Search, highlighting free embedding and reranking with select models.

8/6/2026

Introduces Agent Readiness and Answer Engine Optimization (AEO) tools within the Cloudflare dashboard. Agent Readiness provides diagnostics to check if a site is accessible and discoverable by AI agents, including checks for robots.txt, sitemaps, machine-readable content, and advanced integrations like OAuth discovery and agent-specific protocols. AEO measures how often a site is recommended by AI assistants by inferring industry, probing leading assistants (Claude, GPT) with prompts, and reporting metrics like Citation Rate, Prominence, Mention Rate, and Share of Voice. It also includes an AI Operator Activity view to track bot crawl and referral traffic.

8/5/2026

This post details the internal development and release of Cloudflare OS, a platform designed to enable safe and productive use of AI agents within Cloudflare. It outlines the five guiding principles for AI adoption: customer focus, empowering employees with 'superpowers', human ownership of AI output, context-aware AI, and strict permission controls. It describes the creation of the Cloudflare Engineering Codex for AI-assisted code review and design, and the 'magic AI email bot' approach to identify and automate mundane tasks for non-engineering teams. Finally, it introduces Cloudflare OS v1, a browser-based harness for running skill files and workflows, emphasizing secure access via Cloudflare Zero Trust and ephemeral, auditable environments.

8/5/2026

This post introduces Identity-aware AI Gateway with Cloudflare Access integration and User Insights. Identity-aware AI Gateway allows for authentication with SAML identity providers, policy enforcement, and custom hostnames for AI Gateway access. It adds verified user IDs to request metadata, enabling per-user spend limits and budgeting. User Insights provides behavioral baselining for each account (person or agent) by analyzing session costs against historical data (p95 session cost over 30 days) and an account-level p99 ceiling to detect anomalous spending patterns. It focuses on identifying deviations from normal behavior, flagging potential compromised credentials or rogue agents without blocking actions or deciding intent.

8/5/2026

Introduces Cloudflare OS, an open-source platform for agents, apps, and work. It provides an agent workspace grounded in company context and skills, a new security and governance framework for safe access to internal data and services (including Gatekeepers and resource observation logging), and a platform for building and sharing personal, modifiable apps. The platform allows agents to draw on curated knowledge and work with connected tools and data, generate documents, slides, and spreadsheets, create collaborative apps, and run deterministic workflows. It also supports existing Model Context Protocol (MCP) servers via MCP Server Portals.

8/5/2026

Introduced WriteGuard, a new layer for controlling AI agent interactions with MCP servers. WriteGuard provides shared policy, attribution, and auditing for MCP tools. It defines risk tiers (Read Only, Minimal Impact, Contained Write, Critical) for tools, enabling/disabling them, and configuring labeling for agent attribution. WriteGuard integrates human identity with agent context to attribute actions and generates scrubbed audit events for fleet-wide queryability. Examples demonstrate its application to GitLab tools like `get_merge_request`, `create_mr_note`, and `merge_mr`.

8/4/2026

This post details the development and implementation of an automated issue triage pipeline for the Astro open-source project, leveraging AI agents. It describes the creation of the Flue framework, an open, platform-agnostic system for building durable agents and workflows, and the triagebot-action, a GitHub Action that automates the process of reproducing, diagnosing, and fixing bugs. The contribution highlights the use of isolated AI subagents, state machines driven by issue labels, and the integration with GitHub Actions for transparency and auditability. It also discusses the benefits of this automation, including reducing maintainer burnout and improving code quality by identifying architectural or documentation issues.

8/3/2026

This post details three specific techniques used to optimize the serving of large language models (Kimi and GLM) on Cloudflare Workers AI: 1. Quantizing the KV cache from 16-bit precision (BF16) to 8-bit floating point (FP8) to double the context size and increase concurrent requests. 2. Compressing model weights from 8-bit floating point (FP8) to 4-bit integers (INT4) to reduce checkpoint size and improve decode speed. 3. Implementing KV cache integrity checking with page tags to prevent data corruption from shared cache pages, with a measured cost of under 1% on throughput and latency. The post also highlights the use of SGLang for inference serving and the decision to use different precision/compression levels for prefill and decode phases based on whether they are compute-bound or memory-bound.

7/1/2026

This post introduces the Cloudflare Monetization Gateway, a new capability that allows customers to charge for any resource protected by Cloudflare (web pages, datasets, APIs, etc.) using the x402 protocol and stablecoins. It directly addresses the shift in web business models driven by AI agents, proposing usage-based pricing and micropayments as the new standard. The gateway handles payment verification and enforcement at the edge, simplifying implementation for businesses and enabling new revenue opportunities for creators and API providers.

7/1/2026

This post details the significant shift in internet traffic towards AI-driven consumption and training, with AI crawlers now comprising over 50% of internet traffic. It highlights the decline of traditional referral traffic due to AI summarization and the emergence of a new content licensing market. Cloudflare's role in enabling this market through transparency, control, and scarcity tools is emphasized, along with the challenges posed by mixed-use crawlers like Google's.

7/1/2026

This post introduces initiatives to make AI search smarter by leveraging Cloudflare's network signals to identify fresh and high-quality content, thereby reducing unnecessary crawling by AI search engines. It also details the evolution of 'Pay Per Crawl' to 'Pay Per Use', with experiments involving Ceramic.ai and You.com, to ensure content creators are compensated when their work is used in AI-generated answers. This includes new reporting features for answer engine optimization (AEO).

7/1/2026

This post introduces a significant evolution in Cloudflare's AI integration by providing granular controls for managing AI bot traffic. It moves beyond a simple 'block AI' approach to a pragmatic taxonomy based on bot behavior (Search, Agent, Training) and content usage. New default settings will be implemented, and BotBase, a comprehensive bot database, is launched for Enterprise customers to enhance visibility and control over automated traffic.

7/1/2026

Introduces the Attribution Business Insights dashboard for Cloudflare Bot Management customers. This dashboard provides granular data on bot traffic, specifically differentiating between human, non-AI bot, and AI bot traffic. It offers site-wide and per-bot-operator crawl-to-referral ratios, and classifies AI crawlers into 'Training', 'Search', and 'Agent' categories based on their behavior. The goal is to empower website owners with data to make informed decisions about AI traffic and content access.

6/18/2026

This post details the development of a model-agnostic 'vulnerability harness' for AI-driven security analysis. It describes the evolution from a single-repo 'skill' to a fleet-wide scanning pipeline, emphasizing state management, persistence, and cross-repo reasoning. The harness is divided into a Vulnerability Discovery Harness (VDH) and a Vulnerability Validation System (VVS), using different models for each stage to ensure adversarial validation and model interchangeability. Key components include Recon, Hunt, Validate, Gapfill, Dedup, Trace, and Feedback agents, all orchestrated to manage context exhaustion and ensure robust vulnerability discovery across diverse codebases.

6/17/2026

This post introduces the Cloudflare Agents SDK as a foundational layer for building production-grade AI agents, providing primitives like durable execution, dynamic code execution, and a durable filesystem. It also highlights Flue, a new open-source framework built on the SDK, which simplifies agent development through a declarative approach and integrates with various platforms. The post details how the Agents SDK addresses key challenges for agent harnesses, such as durable execution via Fibers, secure code execution using Code Mode with Dynamic Workers, and persistent file systems with @cloudflare/shell, enabling agents to run reliably at scale.

6/15/2026

This post announces the acquisition of Ensemble AI, a team specializing in AI model compression and efficient inference. Their expertise in techniques like NdLinear and NdLinear-LoRA will be integrated into Cloudflare's AI infrastructure, specifically enhancing Cloudflare Workers AI. The focus is on improving the economics of inference by reducing memory, compute, and deployment overhead for large language models and multimodal architectures, complementing existing efforts like Infire and Unweight.

6/5/2026

Introduces spend controls within Cloudflare AI Gateway, allowing users to set budgets in dollars (not tokens) based on various dimensions like model, provider, or custom attributes (user, team, application). Also announces a closed beta for identity-driven budgets and policies, integrating with Cloudflare Access and existing identity providers to enable per-user and per-team cost attribution and policy enforcement for AI usage. This builds upon existing AI Gateway capabilities by adding granular cost management and visibility.

5/21/2026

This post introduces the integration of Cloudflare CASB with Anthropic's Claude Compliance API. This enables security and compliance teams to monitor Claude usage, including projects, attachments, chat files, chat messages, and artifacts, for misconfigurations and sensitive data without requiring endpoint agents. It leverages the Claude Compliance API to surface security findings directly in the Cloudflare dashboard, allowing for unified triage, assignment, and remediation workflows alongside other SaaS applications. The integration supports both Claude Enterprise and Claude Platform, and can trigger Gateway policies for remediation.

5/18/2026

This post details Cloudflare's experience using Anthropic's Mythos Preview LLM for AI-assisted vulnerability research. It highlights Mythos's capabilities in exploit chain construction and proof generation, contrasting it with general-purpose models. The post also discusses challenges with model refusals in legitimate research, the signal-to-noise problem in vulnerability discovery, and the limitations of using generic coding agents for this task. It introduces the concept of a 'harness' to manage LLM execution for vulnerability discovery, detailing a multi-stage process involving narrow scoping, adversarial review, and parallel task execution.

4/20/2026

This post announces the launch of Cloudflare's 'Agents Week 2026,' marking a significant expansion into the 'agentic cloud.' It details numerous new products and features designed to support AI agents, including compute environments (Sandboxes, Artifacts, Durable Objects in Dynamic Workers), security enhancements (Cloudflare Mesh, Managed OAuth for Access), an agent toolbox (Project Think, voice pipeline, Email Service, AI inference layer, Agent Memory, AI Search, Browser Run), and tools for production deployment (CLI, Agent Lee, Flagship, Registrar API). It also addresses the 'agentic web' with an Agent Readiness score and Redirects for AI Training, and provides a network performance update.

4/20/2026

This post details the architecture and implementation of an AI-powered code review system at Cloudflare. It describes the use of a CI-native orchestration system built around OpenCode, employing specialized AI agents for security, performance, code quality, documentation, release management, and compliance. The system utilizes a plugin architecture for flexibility, processes output via JSONL, and includes mechanisms for real-time processing, error handling, and user feedback to manage LLM thinking time. The core contribution is the engineering approach to integrating LLMs into the critical path of the CI/CD pipeline for automated code review.

4/20/2026

This post details the internal development and deployment of an AI engineering stack at Cloudflare, leveraging existing Cloudflare products such as Cloudflare Access, AI Gateway, and Workers AI. It highlights the architecture for authentication, routing, and inference, including the use of a proxy Worker for centralized control and the integration of open-source models on Workers AI for cost and latency benefits. It also introduces the MCP Server Portal for unified OAuth access and the Agents SDK for stateful agent sessions, showcasing how internal AI adoption drives product development and adoption.

4/17/2026

Introduced isitagentready.com, a tool to score website readiness for AI agents based on discoverability (robots.txt, sitemap.xml, Link Headers), content accessibility (Markdown content negotiation), bot access control (Content Signals, AI bot rules in robots.txt, Web Bot Auth), and capabilities (Agent Skills, API Catalogs, MCP Server Cards). Also launched a new dataset on Cloudflare Radar to track the adoption of AI agent standards across the internet. Detailed the technical implementation of the agent readiness checks and the agent-ready nature of isitagentready.com itself, which exposes MCP server and Agent Skills index.

4/17/2026

This post introduces Agent Memory, a new managed service that provides persistent memory for AI agents. It addresses the challenge of context rot by extracting and storing relevant information from agent conversations, making it available for retrieval without filling up the context window. Agent Memory allows agents to recall what matters, forget what doesn't, and get smarter over time, enhancing their performance in production workloads and enabling new use cases like shared memory across agents and teams.

4/17/2026

This post introduces Flagship, Cloudflare's native feature flag service, specifically designed to support the increasing use of AI in code generation and autonomous agentic workflows. It highlights how feature flags are crucial for safely enabling AI agents to deploy, test, and iterate on code in production by controlling the blast radius. Flagship is built on Cloudflare's infrastructure (Workers, KV, Durable Objects) for edge performance and integrates with the OpenFeature standard for portability.

4/17/2026

Introduced Unweight, a lossless compression system for LLM model weights that reduces model size by 15-22% and saves ~3 GB VRAM. Unweight exploits redundancy in the exponent byte of BF16 weights using Huffman coding, achieving ~30% compression on the exponent stream. It integrates with Cloudflare's Rust-based inference engine and offers four execution pipelines (full decode, exponent-only decode, palette transcode, direct palette) to balance decompression effort and computation complexity based on workload characteristics. This enables running more models on fewer GPUs, making inference cheaper and faster.

4/16/2026

Introduces Cloudflare's AI Platform as a unified inference layer, providing a single API (AI Gateway) to access over 70 models from 12+ providers, including image, video, and speech models. This enables developers to switch models with a one-line code change and manage all AI spend centrally. It also introduces the capability to bring your own models to Workers AI using Cog technology for containerization. The post highlights performance benefits for agents through reduced latency and improved reliability with automatic failover.

4/16/2026

This post details the technical foundation for running extra-large language models on Cloudflare's Workers AI platform. It introduces hardware configurations optimized for agentic use cases, focusing on fast input token processing and tool calling. Key technical contributions include: Prefill Decode (PD) disaggregation for independent scaling and tuning of prefill and decode stages, token-aware load balancing, prompt caching using `x-session-affinity` headers to improve cache hit ratios, KV-cache optimization leveraging Moonshot AI's Mooncake Transfer Engine and Store for multi-GPU sharing and NVMe extension, and speculative decoding using NVIDIA's EAGLE-3 draft model for faster token generation. It also highlights the enhancements to Cloudflare's proprietary inference engine, Infire, including multi-GPU support (pipeline and tensor parallelism) and reduced memory overhead.

4/16/2026

This post introduces Artifacts, a new versioned file system built for agents that speaks Git. It addresses the challenges of scaling source control for AI-driven code generation by enabling programmatic repository creation, management, and access via REST API and Workers API. Artifacts leverages Durable Objects and a custom Zig-based Git server to provide a distributed, versioned storage solution for agent sessions, sandboxes, and any data requiring state tracking, time-travel, and persistence. It also introduces ArtifactFS for faster cloning of large repositories.

4/16/2026

Introduces AI Search, a new primitive for agents that unifies semantic and keyword search. Details hybrid search implementation using vector search and BM25, including tokenizer and fusion configurations. Demonstrates dynamic instance creation for per-agent or per-customer search contexts using `ai_search_namespaces` binding. Provides a detailed example of a customer support agent using AI Search with the Agents SDK, showcasing tool definitions for searching knowledge bases and saving resolutions.

4/15/2026

This post introduces Project Think, a new set of primitives for the Agents SDK designed to build next-generation AI agents. It addresses limitations of current AI agents by enabling durable execution with fibers, sub-agents via Facets for delegation, persistent sessions with tree-structured messages and FTS5 search, and sandboxed code execution using Dynamic Workers and codemode. This significantly enhances the scalability, cost-effectiveness, and robustness of AI agent deployments by leveraging Cloudflare's infrastructure.

4/15/2026

This post introduces an experimental voice pipeline for the Agents SDK, enabling real-time voice interaction with AI agents. It allows developers to add speech-to-text and text-to-speech capabilities to existing agent architectures, leveraging Workers AI for transcription and synthesis. This extends the agent's interaction modalities beyond text, allowing for more natural and conversational user experiences, while maintaining state persistence and WebSocket connectivity.

4/15/2026

This post announces the renaming of 'Browser Rendering' to 'Browser Run' and introduces significant new features specifically designed for AI agents. These include Live View for real-time observation, Human in the Loop for agent assistance, direct Chrome DevTools Protocol (CDP) endpoint for enhanced control and broader language support, MCP client support for AI coding assistants, WebMCP for agent-friendly website interactions, session recordings for debugging, and a substantial increase in concurrent browser limits. These additions position Browser Run as a primary tool for AI agents to interact with the web.

4/15/2026

Introduces Agent Lee, an in-dashboard AI assistant that acts as a new interface to the Cloudflare stack. It leverages Codemode for converting tools into a TypeScript API for LLMs, a Durable Object as a credentialed proxy for sandboxed execution and permission enforcement (read/write classification and explicit approval for writes), and the MCP permission system for querying and executing API requests. Agent Lee also features Generative UI for dynamic UI component generation alongside textual responses and measures quality and safety through evals, feedback signals, and tool call metrics. The vision extends to a platform-wide interface accessible from anywhere and proactive agent capabilities.

4/14/2026

This post introduces a reference architecture for securing enterprise deployments of Model Context Protocol (MCP) and agentic workflows. It details the use of remote MCP servers hosted on Cloudflare's developer platform for better visibility and control, secured by Cloudflare Access for authentication. MCP server portals are introduced for centralized discovery, governance, and data loss prevention. A key contribution is the 'Code Mode' pattern for MCP server portals, which collapses upstream MCP servers into two tools (`portal_codemode_search` and `portal_codemode_execute`) to drastically reduce token costs by enabling on-demand tool discovery and execution within a sandboxed environment powered by Dynamic Workers.

4/13/2026

This post announces the General Availability of Cloudflare Sandboxes and Cloudflare Containers, marking a significant step in enabling AI agents to develop and run code in secure, isolated environments. It details new features like secure credential injection, PTY support for real-time terminals, persistent code interpreters, background processes with live preview URLs, filesystem watching, and snapshots for quick state restoration. The post highlights how these capabilities address key challenges in agentic workloads, such as burstiness, state management, and security, and showcases early adoption by partners like Figma.

4/13/2026

This post introduces outbound Workers for Cloudflare Sandboxes and Containers, enabling programmatic egress proxies for secure connectivity, observability, and flexible authentication for AI agents. It details how these outbound Workers can enforce zero-trust principles, inject credentials dynamically, integrate with other Cloudflare developer platform services (R2, KV), and provide dynamic networking controls, significantly enhancing the security and manageability of agentic workloads.

3/19/2026

This post announces the integration of large language models (LLMs) into Workers AI, starting with Moonshot AI's Kimi K2.5. It details the technical advancements made to support these models, including custom kernels for the Infire inference engine, optimizations for GPU utilization, and the implementation of parallelization techniques. The post also introduces platform improvements for agentic workloads, such as prefix caching with surfaced metrics and discounts, a new `x-session-affinity` header for improved cache hit rates, and redesigned asynchronous APIs for durable inference processing. It highlights significant cost savings achieved by using Kimi K2.5 for internal security review agents.

3/11/2026

Introduced RFC 9457-compliant structured Markdown and JSON error payloads for AI agents, replacing HTML error pages. This provides machine-readable instructions, reduces payload size and token usage by over 98% for common errors like rate limiting, and offers stable fields for agent control flow (e.g., retryable, retry_after, owner_action_required). The implementation covers all 1xxx-class errors and will extend to 4xx/5xx errors, providing a consistent network-wide contract for agentic web interactions.

3/11/2026

This post announces the general availability of AI Security for Apps, a new capability designed to detect and mitigate threats to AI-powered applications. Key contributions include making AI endpoint discovery free for all customers, introducing custom topic detection and custom prompt extraction for enhanced detection accuracy, and expanding collaborations with IBM and Wiz for unified AI security posture management. The post highlights the unique attack surface presented by AI applications and agents, and how AI Security for Apps addresses these risks through discovery, detection, and mitigation.

3/3/2026

This post introduces Cloudy, an LLM-powered explanation layer built into Cloudflare One. Cloudy translates complex machine learning outputs from security detections into human-readable guidance for security teams and end users. It is being extended across Phishnet and API CASB to improve decision making and reduce noise. For Email Security users, Cloudy provides explanations for why messages are flagged, and for CASB, it helps administrators understand the risk and remediation path for SaaS findings. The implementation leverages Cloudflare Workers AI and the global Workers platform for real-time, low-latency summaries. The post details the technical approach for generating these summaries, including the aggregation of structured outputs from multiple detection models and the use of purpose-built prompts for natural language explanations. It also highlights the focus on accuracy and the avoidance of customer data in training.

2/20/2026

Introduced a new Model Context Protocol (MCP) server for the entire Cloudflare API that leverages 'Code Mode' to significantly reduce token usage. This server provides access to over 2,500 API endpoints through just two tools, `search()` and `execute()`, enabling AI agents to interact with Cloudflare services efficiently. Also open-sourced a Code Mode SDK for developers to implement similar patterns.

2/12/2026

Introduced 'Markdown for Agents', a feature that enables real-time conversion of HTML to Markdown for AI agents and crawlers via content negotiation headers. This reduces token usage and processing complexity for AI systems consuming web content. The feature also includes Content Signals to express preferences for content usage by AI.

1/30/2026

This post analyzes the competitive advantage Google gains from its dual-purpose Googlebot, which crawls content for both search indexing and generative AI training. It presents Cloudflare data showing Googlebot accesses significantly more unique URLs than other AI crawlers and highlights that publishers are less likely to block Googlebot due to its search traffic referral importance. The post argues that this asymmetry disadvantages other AI companies and advocates for stronger publisher controls over content usage for AI, critiquing the CMA's proposed remedies as insufficient.

1/29/2026

This post introduces Moltworker, a middleware Worker and adapted scripts that allow running the self-hosted AI agent Moltbot on Cloudflare's Sandbox SDK and Developer Platform APIs. It highlights the advancements in Node.js compatibility within Cloudflare Workers, the integration of AI Gateway for managing AI provider secrets and billing, the use of Sandboxes for secure execution of untrusted code, R2 for persistent storage, and Browser Rendering for web automation, demonstrating how Cloudflare's developer platform can host complex AI applications.

1/15/2026

This post announces the acquisition of Human Native, an AI data marketplace, to enhance Cloudflare's AI capabilities. It highlights Human Native's technology for transforming multimedia content into searchable data for AI training, emphasizing licensed data and fair compensation for creators. The post also discusses the strain on the internet's economic model due to AI crawls and introduces Cloudflare's AI Crawl Control and Pay Per Crawl tools. It announces the development of an AI Index for structured content updates via a pub/sub model and the establishment of the x402 Foundation with Coinbase to enable machine-to-machine transactions for digital resources, building on the foundation for new economic models in the age of AI.

12/15/2025

This post contributes to the AI thread by providing concrete data on the evolving landscape of Generative AI services, specifically highlighting the rise of ChatGPT rivals like Claude, Perplexity, and Gemini. It details their growth, user adoption patterns (weekday vs. weekend), and the emergence of new players like DeepSeek, showcasing the dynamic nature of AI service popularity and its impact on information consumption.

12/1/2025

This post announces the acquisition of Replicate by Cloudflare, highlighting Replicate's contributions in packaging ML models with Cog and providing a platform for running them as API endpoints. It emphasizes the evolution of the AI stack beyond just model inference to include microservices, object storage, caching, and databases. The integration with Cloudflare's network, Workers, R2, and Durable Objects is expected to enable new capabilities like running fast models on the edge and model pipelines on Workers.

11/25/2025

This post details the integration of the FLUX.2 open-weight image generation model into Cloudflare Workers AI. It highlights the model's capabilities in generating photorealistic images with physical world grounding and character consistency. The post also describes technical implementation details, including the use of multipart form data for multiple image inputs and output images up to 4 megapixels, and provides code examples for both curl and Workers AI Binding integration. It further explains advanced control mechanisms like JSON prompting and specifying hex codes for color customization.

11/17/2025

This post announces the acquisition of Replicate by Cloudflare, significantly expanding Cloudflare's AI capabilities. It integrates Replicate's extensive model catalog (50,000+ models) and developer community into Cloudflare's Workers AI platform. Key technical advancements include enabling fine-tuning and custom model deployment on Workers AI, leveraging Replicate's Cog tool for seamless model packaging. The integration also enhances the AI Gateway for unified observability, prompt management, A/B testing, and cost analytics across various inference providers. It also highlights the synergy with existing Cloudflare AI services like Vectorize, R2, Agents, and Workflows for building comprehensive AI applications.

10/24/2025

Introduces partnerships with Visa and Mastercard to secure agentic commerce, leveraging Web Bot Auth and HTTP Message Signatures. Details the implementation of Visa's Trusted Agent Protocol and Mastercard's Agent Pay, which use cryptographic signatures to authenticate AI agents for browsing and payment. Outlines the verification checks performed by Cloudflare for these protocols and mentions future integration with Cloudflare's Agent SDK and managed rulesets for WAF.

9/26/2025

Introduces AI Index, a new system for creating AI-optimized search indexes for customer domains, enabling discoverability and monetization of content for AI builders. It also introduces the Open Index, an aggregated layer of participating AI Indexes. The system leverages existing AI Search technology and introduces new APIs like MCP server, LLMs.txt, and pub/sub connections for real-time content updates. It integrates with AI Crawl Control and Pay per crawl for access management and monetization.

9/26/2025

This post introduces 'Code Mode' for the Model Context Protocol (MCP), a new approach to enabling AI agents to interact with external tools. Instead of directly calling tools, LLMs are now prompted to write TypeScript code that calls a TypeScript API representation of the MCP tools. This significantly improves the agent's ability to handle a larger number of complex tools and chain multiple tool calls efficiently, by leveraging the LLM's extensive training on real-world code. It also highlights the uniformity and discoverability benefits of MCP, even when used in a code-generation context.

9/25/2025

This post details Cloudflare's strategy for enabling AI sovereignty through its platform, focusing on infrastructure accessibility at the edge for AI inference, inclusivity via serverless models and a free tier, control over data through edge security tools, and fostering multi-modal, dynamic markets with model-agnostic offerings like Workers AI and AI Gateway. It highlights the availability of over 50 open-source models on Workers AI and introduces specific regional AI innovation initiatives by offering locally-developed models from India (IndicTrans2), Japan (PLaMo-Embedding-1B), and Southeast Asia (SEA-LION v4-27B) on the Workers AI platform, providing concrete examples of API usage for these models.

9/24/2025

Introduced the Content Signals Policy, a new addition to robots.txt that allows website operators to express preferences for how their content can be used after access, specifically defining signals for 'search', 'ai-input', and 'ai-train'. This policy aims to address the free-rider problem and the choice between open access and data lockdown for AI training and usage.

9/24/2025

This post proposes five responsible AI bot principles: public disclosure, self-identification (emphasizing cryptographic verification via Web Bot Auth), declared single purpose, respecting preferences, and acting with good intent. It details the rationale behind each principle and provides examples of current practices and areas for improvement, aiming to foster a more transparent and accountable AI ecosystem.

9/23/2025

This post introduces a new platform capability for per-customer behavioral anomaly detection to combat sophisticated AI-driven bot threats, particularly for AI model training. It details how Cloudflare leverages global bot fingerprinting, client-side information from challenge solves to detect residential proxies, and establishes dynamic, per-customer baselines of normal traffic to identify and act upon anomalies invisible to global systems. This significantly enhances bot management by providing hyper-personalized security tailored to each customer's unique traffic patterns.

9/23/2025

Introduced Cloudflare Application Confidence Scorecards as a new AI Security feature within the Cloudflare One SASE platform. This feature automates the evaluation of generative AI and SaaS applications by assigning scores based on security, privacy, data management, and AI-specific risks. The system involves a rubric for scoring (Application Posture Score and Gen-AI Posture Score), automated internet crawling and parsing of public documents, and human auditing for accuracy. The scores are integrated into the Application Library for visibility and will soon drive enforcement policies via Gateway and DLP.

9/23/2025

This post introduces VibeSDK, an open-source AI vibe coding platform that enables one-click deployment of end-to-end AI development platforms. It details the integration of LLM models via Agents SDK, isolated development environments using Cloudflare Sandboxes, scalable deployment on Cloudflare Workers, observability and caching through AI Gateway, and project export to Cloudflare accounts or GitHub. The post also outlines the technical steps involved in building such a platform, including sandbox creation, code generation, preview URL exposure, and deployment to Workers for Platforms.

9/23/2025

This post introduces the integration of Cloudflare's Bot Management and AI Crawl Control services into Project Galileo, offering free protection against AI crawlers for journalists and non-profits. It details how AI Crawl Control identifies and categorizes AI crawlers, provides data on their activity, and allows website owners to control access, enabling content protection and negotiation for fair compensation.

9/23/2025

Introduces the x402 protocol for machine-to-machine payments, enabling AI agents to pay for resources. Proposes a new deferred payment scheme for x402 to support scenarios like pay-per-crawl and batch settlements. Integrates x402 support into the Agents SDK and MCP servers, demonstrated with a playground and example code for agentic payments and tool usage.

9/3/2025

This post summarizes Cloudflare's AI Week 2025 announcements, detailing new features and product enhancements across four core areas: Securing AI environments and workflows (AI prompt protection, Shadow AI detection, Application Confidence Score, CASB scanning, MCP Server Portals, SASE best practices), Protecting original content from misuse by AI (AI crawler analysis, cryptographic agent recognition, NLWeb/AutoRAG, AI Crawl Control, AI bot traffic analysis), Helping developers build world-class, secure, AI experiences (AI Gateway enhancements, Infire inference engine, Workers AI expansion with Leonardo.Ai and Deepgram, Omni GPU optimization, AI Avenue resources, Firewall for AI, real-time voice agents), and Making Cloudflare better for you with AI (image segmentation model evaluation, Cloudy for threat analysis and email detection summaries, AI-powered network troubleshooting).

8/29/2025

This post details the integration of Cloudflare's AI agent, Cloudy, with its security analytics functionality and introduces a new conversational interface for faster root cause analysis and mitigation. It also describes the integration of Cloudy with the Cloudforce One Threat Events platform, allowing users to query global threat data in natural language. The technical implementation leverages the Agents SDK and Workers AI, with a commitment to not training on customer data.

8/29/2025

Introduces "Cloudy Summaries", an AI-powered agent extension to Cloudflare One, designed to translate complex, multi-signal email detection logic into clear, human-readable explanations for Security Operations Center (SOC) teams. Addresses the "hallucination problem" in LLMs by implementing Retrieval-Augmented Generation (RAG) using detection dataset corpus and enhancing model context for accuracy. This feature aims to reduce investigation time and improve security posture by providing SOC teams with actionable insights into why emails were blocked.

8/29/2025

This post introduces detailed analysis of AI bot crawl activity, categorizing it by purpose (training, search, user actions) and tracking the evolution of specific AI crawlers (GPTBot, ClaudeBot, Bytespider, etc.). It quantifies the 'crawl-to-refer' ratio, highlighting the imbalance between AI crawling for training and the limited referral traffic back to content creators. It also analyzes the impact of AI Overviews on referral traffic from search engines like Google.

8/28/2025

This post introduces new capabilities to the AI Insights page on Cloudflare Radar: industry-focused AI bot traffic data and a breakdown of AI bot traffic by purpose (Training, Search, User action, Undeclared). It details how these insights are presented, including crawl/refer ratios, traffic trends by purpose and industry, and how users can explore this data via Data Explorer with filters for vertical and industry.

8/28/2025

This post details the evaluation of various image segmentation models (U2-Net, IS-Net, BiRefNet, SAM) for background removal functionality. It explains the technical concepts of image segmentation, salient object detection, and the metrics used for evaluation (Intersection over Union, Dice coefficient, pixel accuracy). The post introduces background removal as a new feature leveraging Workers AI.

8/28/2025

Introduces NLWeb, an open standard for natural-language queries on websites, and integrates it with Cloudflare's AutoRAG. AutoRAG is enhanced with website crawling capabilities to index dynamic content from websites into R2 and Vectorize. A Cloudflare Worker is deployed to implement the NLWeb protocol, providing a conversational UI and an MCP server endpoint for AI agents. The AutoRAG architecture is updated with JobManager and FileManager Durable Objects to handle website crawling and improve scalability for parallel file processing.

8/28/2025

Introduces 'signed agents' as a new classification for AI agent traffic, extending the verified bots program. This involves using HTTP message signatures (Web Bot Auth) to cryptographically verify agent requests, allowing website owners to granularly control access. The post details the policy for signed agents, the submission process via the Cloudflare dashboard, and the integration of this classification into security rules and Cloudflare Radar. It also highlights partnerships with initial signed agents like ChatGPT agent, Goose, Browserbase, and Anchor Browser.

8/28/2025

Introduces AI Crawl Control (formerly AI Audit) to general availability, rebranding it to emphasize control over AI crawler access. Details the use of HTTP 402 ('Payment Required') status codes to facilitate licensing agreements between content creators and AI crawlers, allowing customization of messages for communication and commercial terms. Outlines plans for future enhancements to include structured data on content value, freshness, and licensing terms within 402 responses, and mentions the ongoing beta of 'pay per crawl' for automated monetization.

8/27/2025

Introduced unified billing for AI model usage across multiple providers, secure API key management via Secrets Store integration, dynamic routing capabilities for AI traffic, and enhanced security with Data Loss Prevention (DLP) scanning within the AI Gateway Firewall.

8/27/2025

Introduced Omni, an internal platform for running and managing AI models on Cloudflare's edge nodes. Omni enables spawning multiple models from a single control plane, implements lightweight process isolation with filesystem separation, and over-commits GPU memory using a CUDA stub library and unified memory mode to run more models on a single GPU. It also addresses Python memory allocation issues by using FUSE to create a virtual /proc/meminfo reflecting per-model limits and leverages uv for Python virtual environments.

8/27/2025

This post details the development of Infire, a new LLM inference engine written in Rust, designed to maximize memory, network I/O, and GPU utilization for Cloudflare's edge network. It addresses limitations of existing solutions like vLLM by enabling co-hosting of multiple models on a single GPU, secure execution without additional virtualization layers, and optimized performance for dynamic workloads. Key technical contributions include efficient model loading with Page Locked memory and CUDA asynchronous copies, parallel kernel compilation, an OpenAI compatible HTTP server, and a batcher employing continuous batching with chunked prefill to optimize matrix operations and KV cache utilization.

8/27/2025

This post introduces the integration of closed-source partner AI models into Workers AI, specifically Leonardo.Ai's image generation models (@cf/leonardo/phoenix-1.0 and @cf/leonardo/lucid-origin) and Deepgram's speech-to-text (@cf/deepgram/nova-3) and text-to-speech (@cf/deepgram/aura-1) models. It details their performance characteristics, provides API examples for both REST and AI binding usage, and highlights how these models can be combined with other Cloudflare developer products (Workers, R2, Images, Realtime) to build comprehensive AI applications like image generation platforms and real-time voice agents. WebSocket support for Deepgram models is also announced.

8/26/2025

Introduces Cloudflare MCP Server Portals, a new capability within Cloudflare One designed to secure connections between Large Language Models (LLMs) and applications via the Model Context Protocol (MCP). This feature provides centralized policy enforcement, comprehensive visibility and logging, a curated AI user experience based on least privilege, and simplified user configuration for MCP servers. It aims to address security risks such as prompt injection, supply chain attacks, privilege escalation, and data leakage associated with unsecured AI ecosystems.

8/26/2025

This post introduces new AI Security Posture Management (AI-SPM) features built into Cloudflare's SASE platform to address the security challenges of generative AI adoption. These features include shadow AI reporting for visibility into unsanctioned AI use, confidence scoring for AI providers, AI prompt protection against malicious inputs, out-of-band API CASB integrations for detecting misconfigurations, and tools to secure Model Context Protocol (MCP) deployments. The post outlines best practices for developing an AI Security Strategy and leveraging Cloudflare's SASE platform for visibility, risk management, and data protection in the context of AI.

8/26/2025

This post introduces the integration of unsafe content moderation directly into Cloudflare Firewall for AI, leveraging Llama Guard. This feature provides real-time protection for Large Language Models (LLMs) at the network level by detecting and blocking harmful prompts or topics before they reach the model. It offers a model-agnostic approach, unified detection, analytics, and topic enforcement, enhancing AI security offerings and addressing OWASP Top 10 LLM risks.

8/26/2025

Introduces API-based Cloud Access Security Broker (CASB) integrations for ChatGPT, Claude, and Gemini. These integrations enable agentless scanning for security posture, misconfigurations, Data Loss Prevention (DLP) matches, data exposure, and compliance risks within these generative AI tools. Specific detections include capability activation, external exposure of GPTs, secrets/keys/invites, and sensitive content via DLP for ChatGPT; secrets/keys/invites and sensitive content via DLP for Claude; and identity, MFA, and license hygiene for Gemini within Google Workspace.

8/26/2025

Introduces Cloudflare Application Confidence Scores, a new feature within the AI Security Posture Management (AI-SPM) suite. This feature provides automated assessment of SaaS and Gen AI applications based on a transparent, objective rubric. The rubric evaluates regulatory compliance (SOC 2, GDPR, ISO 27001), data management practices (retention windows, third-party sharing), and security controls (MFA, RBAC, session monitoring). It also includes a Gen-AI Confidence Score specifically for AI-related risks such as model training on user input and output safety. The scoring system is automated with human oversight and allows for vendor appeals. Scores are provided for different account tiers of applications, recognizing varying risk profiles.

8/25/2025

Introduces AI prompt protection as a new capability within Cloudflare's Data Loss Prevention (DLP) product, integrated into Cloudflare One. This feature aims to secure sensitive data entered into generative AI tools by providing prompt detection, topic classification (content and intent), guardrails for granular policy creation, and auditable logging of prompts and responses. It supports specific AI tools like Google Gemini, ChatGPT, Claude, and Perplexity, and leverages Kivera's acquisition for operation mapping to analyze API interactions.

8/24/2025

This post announces Cloudflare's AI Week 2025, focusing on four core areas: securing AI environments and workflows, protecting original content from AI misuse, helping developers build secure AI experiences, and integrating AI into Cloudflare's own products. Key announcements include new controls for Shadow AI and unauthorized AI usage, establishing 'paved paths' for AI tooling, protecting AI models from attacks, expanding Crawl Control for content creators, and enhancing the AI Gateway with advanced caching, observability, and framework integrations. The post also highlights the integration of AI into Cloudflare's core services for threat detection and network optimization.

8/20/2025

This post details the technical implementation of AI face cropping for Cloudflare Images. It describes the migration from a CPU-based prototype to a GPU-based implementation using Workers AI, addressing memory leaks for large-scale use. The post explains the selection of the RetinaFace model based on performance on the WIDERFACE dataset, speed, and model size. It also details the inference process, including image resizing for efficiency, the construction of bounding boxes for detected faces, and the calculation of the outer bounding box and focal point for cropping. The privacy considerations of using face detection rather than facial recognition are also highlighted.

8/5/2025

This post announces the integration of OpenAI's new open-weight models (120B and 20B parameters) onto Cloudflare Workers AI. It details the technical specifications of these models, including their Mixture-of-Experts (MoE) architecture and FP4 quantization. The post also highlights the implementation of OpenAI's Responses API and the upcoming Chat Completions API compatibility. A key technical contribution is the integration of Code Interpreter with Cloudflare Sandboxes, enabling stateful code execution for AI-generated code within a secure, ephemeral container environment.

7/1/2025

Introduces two new tools for website owners to control AI bot access to their content: a managed robots.txt file service and an option to block AI bots only on monetized portions of a site. This addresses the shift in the crawler relationship from symbiotic to extractive, evidenced by high crawl-to-referral ratios for AI bots. The post also provides updated data on AI bot traffic and usage of Cloudflare's existing AI bot blocking features, highlighting the underutilization of robots.txt for AI crawlers and the increasing adoption of Cloudflare's AI bot management tools.

7/1/2025

This post analyzes the evolving landscape of web crawlers, with a specific focus on the rise of AI crawlers. It presents data on the changing market share of AI bots like GPTBot, ClaudeBot, and Meta-ExternalAgent, and highlights the significant growth of Googlebot. The post also discusses the implications of AI crawling for content creators and the tools Cloudflare is developing to manage and enforce crawler policies, such as AI Audit.

7/1/2025

Introduces 'pay per crawl' functionality, enabling content owners to charge AI crawlers for access using HTTP 402 Payment Required status codes and authenticated requests. This feature integrates with existing web infrastructure, WAF, and bot management, and includes mechanisms for publishers to define pricing and for crawlers to authenticate and pay via HTTP headers.

7/1/2025

Introduced a new metric on Cloudflare Radar's AI Insights page to measure the ratio of AI bot crawl requests to actual referral traffic. This metric helps content providers understand how AI bots are consuming their content versus driving traffic. The post details the methodology for calculating this ratio, including the use of User-Agent strings and Referer headers, and presents observations on crawl-to-referral ratios for various AI platforms and search engines, highlighting trends and changes over time.

6/18/2025

This post introduces `use-mcp`, a React library that simplifies connecting React applications to remote Model Context Protocol (MCP) servers, handling transport, authentication, and session management in minimal code. It also open-sources the AI Playground, a chat interface built on Workers AI that uses `use-mcp` to interact with MCP servers, demonstrating dynamic tool discovery, connection management with retries, and OAuth 2.1 authentication flows. The post details how the AI Playground leverages `use-mcp` and Workers AI for tool invocation and provides debugging capabilities.

5/1/2025

This post introduces Cloudflare's first publicly available remote Model Context Protocol (MCP) servers, enabling AI clients to connect to Cloudflare's services. It details the functionality of various MCP servers, including those for Cloudflare Documentation, Workers Bindings, Workers Observability, Container sandboxing, Browser Rendering, Radar data, Logpush job health, AI Gateway logs, AutoRAG documents, Audit Logs, DNS Analytics, Digital Experience Monitoring, and Cloudflare One CASB. It also highlights the collaboration with Anthropic and other companies building MCP servers on Cloudflare.

5/1/2025

This post details Cloudflare's advancements in simplifying the deployment and management of remote Model Context Protocol (MCP) servers. Key contributions include: one-click deployment of pre-built MCP servers, support for the new Streamable HTTP transport alongside SSE, enabling Python as a development language for MCP servers on Cloudflare, and providing improved documentation, starter templates, and best practices for building MCP servers. It also highlights how partner companies (Asana, Atlassian, Intercom) are leveraging Cloudflare's infrastructure to build and deploy their MCP servers.

4/30/2025

This post introduces two key advancements for the Model Context Protocol (MCP) within Cloudflare's AI ecosystem: the new Streamable HTTP transport, which simplifies communication between AI agents and services by using a single HTTP endpoint for sending and receiving responses, and first-class Python language support for building MCP servers on Cloudflare Workers. This enhances backward compatibility, reduces complexity, and broadens the developer base for AI agent development.

4/11/2025

This post highlights how startups are leveraging Cloudflare's AI capabilities to build scalable AI agent platforms. Lamatic AI uses Workers, Queues, and Workers KV for a no-code visual builder for AI agents, achieving millions of serverless requests per month. Skyward AI uses Durable Objects, Workers, and AI Gateway to automate compliance operations with AI-driven agents, achieving sub-100ms response times and simplifying agent communication with the new AI agents framework. Both examples showcase Cloudflare's infrastructure as a backbone for AI innovation, emphasizing performance, scalability, and developer experience.

4/11/2025

Introduced speculative decoding and prefix caching to significantly speed up inference times (2-4x) for models like Llama 3.3 70b. Launched an asynchronous batch API to handle large workloads more reliably, preventing immediate errors due to capacity and guaranteeing fulfillment. Expanded LoRA support to 8 models with larger ranks (up to 32) and file sizes (up to 300 MB). Added new models to the catalog and refreshed the dashboard for improved usability.

4/9/2025

Introduces Cloudflare Secrets Store (Beta) as a centralized platform for managing secrets like API tokens and credentials across Cloudflare products, starting with Cloudflare Workers. Details the security architecture involving DEKs and KEKs, RBAC for access control, and integration with Quicksilver for network-wide propagation. Highlights future plans to support more products like Cloudflare Access, Transform Rules, and AI Gateway, and to allow binding to the store itself and multiple secret stores per account.

4/7/2025

Introduces AutoRAG, a fully managed Retrieval-Augmented Generation (RAG) pipeline powered by Cloudflare's developer platform. AutoRAG simplifies RAG pipeline creation by handling data ingestion, chunking, embedding, vector storage in Vectorize, semantic retrieval, and response generation using Workers AI. It features continuous background indexing and monitoring of data sources. The post details the indexing process (file ingestion from R2, Markdown conversion via Workers AI, chunking, embedding, Vectorize storage) and the querying process (query reception, optional query rewriting, query embedding, Vectorize search, metadata/content retrieval, response generation). It also provides a tutorial on using the Browser Rendering API with a Worker to fetch web pages and store them in R2 for AutoRAG.

4/7/2025

This post significantly advances Cloudflare's AI agent capabilities by introducing new Agents SDK features for building remote MCP clients, enabling AI agents to connect to and interact with external services. It also integrates BYO OAuth providers (Stytch, Auth0, WorkOS) for authentication and authorization of MCP servers, introduces hibernation for stateful remote MCP servers to manage costs, and makes Durable Objects available on the free tier, a key component for agent development. The post also announces the general availability of Workflows for production-ready actions and AutoRAG for easier context-aware AI integration, alongside a new landing page for agents.

4/6/2025

This post announces the availability of Meta's Llama 4 models (specifically Llama 4 Scout) on the Cloudflare Workers AI platform. It details the technical advancements in Llama 4, including its Mixture of Experts (MoE) architecture for efficient inference and its natively multimodal capabilities (understanding text and images). The post highlights how Workers AI provides a serverless solution for hosting these large models, abstracting away infrastructure complexities and enabling developers to easily integrate Llama 4 into their applications with large context windows.

3/25/2025

This post introduces the ability to build and deploy remote Model Context Protocol (MCP) servers to Cloudflare, making AI models and their associated tools accessible over the internet. It highlights new features like workers-oauth-provider for simplified authorization, McpAgent for remote transport, mcp-remote for client compatibility, and an AI playground as a remote MCP client. The post emphasizes the shift from local MCP servers to accessible remote ones, enabling broader consumer use cases and addressing the challenges of authentication and authorization for AI agents interacting with external services.

3/21/2025

Introduces AI-powered context analysis to Cloudflare's Data Loss Prevention (DLP) solution, leveraging Workers AI for text embedding and Vectorize for vector database storage. This feature aims to reduce false positives by analyzing the context of data matches and learning from customer feedback. It also details the integration with Cloudflare Queues for asynchronous indexing and discusses latency implications and future multilingual support.

3/20/2025

Introduces Cloudy, Cloudflare's first AI agent, designed to simplify complex configurations within WAF Custom Rules and Cloudflare One Gateway policies. Cloudy leverages Workers AI and LLMs to analyze policy and rule data, providing summaries and actionable recommendations. Specific functionalities include identifying redundant/conflicting/disabled rules, optimizing execution order for WAF Custom Rules, and summarizing Gateway policies for clarity. The underlying mechanism involves ingesting policy/rule data, enriching it with product documentation and translated list items, and using LLMs with specific output instructions and a focus on accuracy to avoid hallucinations.

3/19/2025

Introduces 'Cloudflare for AI', a suite of tools for adopting, deploying, and securing AI technologies at scale. Highlights Workers AI and AI Agents SDK for building and deploying AI applications at the edge, R2 for storing AI training data, AI Gateway for visibility into cost, usage, and performance, and Firewall for AI for security, including prompt injection prevention, PII protection, and exploit mitigation. Also introduces AI Crawl Control for monitoring and blocking AI crawlers accessing content for training data.

3/19/2025

This post introduces the concept of cryptographic watermarking for AI-generated content as a potential solution to challenges in identifying AI artifacts and ensuring their provenance. It discusses the goals of robustness, undetectability, and unforgeability, contrasting deep learning-based watermarks with cryptographic approaches. The post details the framework of pseudorandom error-correcting codes, including key generation, encoding, and decoding algorithms, as a promising direction for developing robust and secure AI watermarks.

3/19/2025

This post details the integration of a new AI model into Cloudflare's Page Shield product. This model uses Graph Neural Networks (GNNs) to analyze JavaScript syntax trees and detect specific malicious intents (Magecart, crypto mining, malware) with greater precision than previous methods. This represents a significant advancement in Cloudflare's use of AI for proactive client-side security.

3/19/2025

This post introduces Firewall for AI, an inline security solution that protects user-facing LLM-powered applications. It focuses on three key capabilities: discovering LLM-powered endpoints using heuristics like request path signatures, server-sent events, and origin response bitrate; detecting Personally Identifiable Information (PII) in prompts using a hybrid approach of regex and Named Entity Recognition (NER) models deployed via Workers AI with Presidio; and mitigating risks by integrating with Cloudflare's WAF for policy enforcement. The post details the technical challenges and solutions for discovering LLM endpoints and the architecture for PII detection and mitigation.

3/19/2025

Introduced AI Labyrinth, a new mitigation approach that uses AI-generated content to slow down, confuse, and waste the resources of AI Crawlers and other bots that don't respect "no crawl" directives. This involves using Workers AI with open source models to pre-generate HTML pages, storing them in R2, and integrating them as hidden links via custom HTML transformation. AI Labyrinth also acts as a next-generation honeypot, identifying and fingerprinting bad bots by analyzing which crawlers follow these hidden pathways, feeding data into machine learning models for improved bot detection.

3/10/2025

This post details the evolution of Generative AI services throughout 2024 and early 2025, analyzing their traffic growth, popularity, and the associated cyberattack landscape from Cloudflare's perspective. It leverages aggregated data from the 1.1.1.1 DNS resolver and Cloudflare's customer base to identify key trends, including the sustained dominance of ChatGPT, rapid traffic growth across Generative AI services, the emergence of new players like DeepSeek and Grok/xAI, and a significant increase in blocked attack requests, particularly DDoS campaigns targeting AI chatbots and image-generation sites. The post also provides insights into the global distribution of Generative AI visitors and the specific traffic patterns of various popular AI services.

2/26/2025

Introduces Guardrails in AI Gateway, a new feature that leverages Llama Guard on Workers AI to provide content moderation for AI prompts and responses. This feature aims to ensure consistent and safe AI application deployment by allowing developers to define evaluation scopes and hazard categories, with options to flag or block inappropriate content. It addresses challenges of model inconsistency and lack of visibility into unsafe content.

2/25/2025

This post introduces the `agents-sdk` framework for building AI agents on Cloudflare Workers, along with updates to Workers AI to support structured outputs, tool calling, and longer context windows. It details the concept of AI agents and how they differ from traditional automation and co-pilots, providing examples and code snippets for building and deploying agents. The post emphasizes Cloudflare's commitment to being the ideal platform for AI agent development.

2/4/2025

This post introduces the new 'AI Insights' page on Cloudflare Radar, significantly expanding the platform's ability to track and analyze AI trends. It adds new metrics for AI bot and crawler traffic, provides real-time popularity rankings of Generative AI services based on DNS resolver traffic, analyzes how content providers use robots.txt directives to manage AI bot access, and offers insights into the popularity of models and tasks on Workers AI. This enhances Cloudflare's role in providing visibility into the dynamic AI landscape.

2/3/2025

This post introduces the integration of the Coalition for Content Provenance and Authenticity (C2PA) standard into Cloudflare Images. This feature allows content creators and publishers to preserve the provenance chain of images, including creation details and subsequent edits, across the Cloudflare network. When images are transformed by Cloudflare Images (e.g., resizing, format changes), the existing Content Credentials are preserved, and Cloudflare cryptographically signs any additional transformations. This ensures that end-users can verify the full provenance chain using open-source verification services. The implementation involves embedding Content Credentials using JUMBF and cryptographically signing assertions with Cloudflare's end-entity certificate, which is part of a Trust List.

1/14/2025

This post details Cloudflare's proactive approach to reducing entire classes of vulnerabilities, specifically injection and secrets, aligning with CISA's 'Secure by Design' pledge. It highlights the implementation of custom rulesets, 'build break' enforcement in CI/CD pipelines, and developer enablement to prevent vulnerabilities at the source, leading to significant reductions in secrets (79%) and potential SQL/code injection vulnerabilities (44%). This demonstrates a concrete application of secure-by-design principles to enhance product security.

12/24/2024

This post details an experiment using Cloudflare Workers AI to translate automatically generated captions from Cloudflare Stream. It explores parsing VTT files, leveraging the 'm2m100-1.2b' translation model, and addresses challenges with translation quality due to fragmented captions by implementing a sentence consolidation pre-processor. The experiment demonstrates the feasibility of using Workers AI for multilingual captioning and informs future product development for Cloudflare Stream.

12/20/2024

This post introduces the capability to build and deploy MCP servers on Cloudflare Workers, enabling LLMs like Claude to interact with custom services. It details the use of the `workers-mcp` tooling, which simplifies the process by handling MCP standard translation and using JSDoc for tool definition. An example of adding image generation functionality using Workers AI is provided, showcasing the ease of integrating custom tools.

12/20/2024

This post details the integration of Cloudflare Calls with OpenAI's Realtime API to enable multimodal, real-time interaction with AI applications. It explains how Cloudflare Calls leverages WebRTC to overcome the latency limitations of WebSockets for audio and video streams, allowing multiple users to interact with AI simultaneously. The post provides a code example demonstrating how to connect human and AI sessions by adding AI audio tracks to human sessions and vice-versa, facilitating bidirectional communication.

12/10/2024

Introduces Robotcop, an extension of AI Crawl Control, that allows customers to enforce their robots.txt policies for AI bots at the network level via WAF rules. This moves beyond voluntary compliance to programmatic enforcement of content access rules for AI crawlers.

12/9/2024

This post analyzes the rise of Generative AI services, specifically highlighting ChatGPT's significant growth and its impact on overall internet rankings, as well as the emergence of other AI tools like GitHub Copilot. It provides data-driven insights into user interaction trends with these AI services throughout 2024, demonstrating their increasing integration into daily online activities.

11/19/2024

Introduced a WebSocket API for AI Gateway, enabling persistent, real-time communication with AI providers. Implemented authentication using Cloudflare API tokens via `cf-aig-authorization` header and `sec-websocket-protocol` for browser compatibility. Leveraged Durable Objects for message handling and introduced the `eventId` field in the Universal format to correlate asynchronous messages in streaming scenarios.

10/24/2024

This post details the evolution of AI Gateway's logging capabilities, addressing the critical challenge of storing billions of logs indefinitely. It describes the transition from a 30-minute log retention limit to persistent storage by migrating request bodies to R2 and utilizing Durable Objects with SQLite for indexed logs. The introduction of sharding by account ID and gateway name, along with an Account Manager Durable Object for entitlement and usage tracking, significantly scales log storage capacity and management. This enables advanced features like AI evaluations and Human-in-the-Loop feedback by providing persistent, searchable historical data.

9/27/2024

This post introduces the AI Assistant for WAF Rule Builder, allowing users to generate WAF rules using natural language prompts. It also adds AI bot traffic insights to Cloudflare Radar, providing visibility into the activity of AI bots and crawlers. Furthermore, it enhances AI bot protection by expanding the 'block AI bots' feature to include more AI bot signatures and enabling dynamic updates for the AI bot rule, moving towards a managed rule framework. The post also details the impact of these new fingerprints on Cloudflare's ML model and observes a decline in AI bot traffic, hypothesizing that Cloudflare's blocking features are discouraging them.

9/26/2024

This post announces significant upgrades to Cloudflare's AI platform for Birthday Week 2024. Key contributions include: 1. Workers AI upgrades: more powerful GPUs enabling faster inference and support for larger models (e.g., Llama 3.1 70B), expanded GPU footprint to over 180 cities, improved performance metrics (80+ TPS for 8b models, 300ms TTFT), support for larger context windows (128K), Day 0 support for Meta's Llama 3.2 models, and the introduction of 'Run Any* Model' for dynamic model loading (closed beta). A technical blog post details KV cache compression, speculative decoding, and new hardware design. 2. Workers AI pricing: transition from 'neurons' to a unit-based pricing model (tokens, image resolution/steps, audio seconds) with updated free tier allocations. 3. AI Gateway enhancements: introduction of persistent logs (open beta) for up to 10 million logs per gateway with enhanced logging interface (cost, duration, human feedback, custom metadata), Logpush integration for exporting logs, and AI evaluations (open beta) using human-in-the-loop feedback for scoring model performance, speed, and cost. 4. Vectorize: General Availability (GA) with expanded index sizes and faster queries.

9/26/2024

Introduced two performance optimization techniques for Workers AI: KV cache compression using PagedAttention to increase throughput by up to 8x and speculative decoding (specifically prompt-lookup decoding) to improve generation speed by up to 40% for Llama-3.1-8B-instruct and up to 70% for the 70B model. Also announced upgraded hardware supporting newer GPUs and larger models like Meta Llama 3.2 11B and Llama 3.1 70B, offering 2-3x throughput for these models.

9/23/2024

This post introduces new tools for site owners to audit and control how AI models access their content. It includes an analytics view of AI services crawling sites, a one-click option to block all AI crawlers, and filters in the WAF to allow specific AI providers or bot types. It also previews a monetization feature for content usage by AI models.

9/12/2024

This post details how Cloudflare is using statistical machine learning techniques, specifically sequence learning with variable order Markov chains, to protect APIs against sequential abuse. It introduces the concept of API sessions and explains how these models help distinguish malicious from benign API request sequences, building upon previous work in API sequence analytics.

7/25/2024

This post details performance optimizations for Cloudflare's WAF ML product, specifically the WAF Attack Score. It describes the system design, identifies bottlenecks in feature extraction (HashMap lookups and regex replacements), and presents an optimization using the Aho-Corasick algorithm to replace HashMap lookups, resulting in a 47.84% reduction in processing time for long inputs and a 1.64x speedup. Further optimizations for model inference are also discussed.

7/23/2024

This post announces the availability of Meta's Llama 3.1 8B model on Cloudflare Workers AI, marking an expansion of the platform's supported open-source models. It highlights the model's features, including higher precision (bfloat16), function calling, and multilingual support, and demonstrates its integration with Workers AI's serverless inference platform and embedded function calling capabilities. The post also provides code examples for using the model and mentions the Workers AI Playground.

6/27/2024

Introduces embedded function calling in Workers AI, allowing LLM inference and function execution within the same environment to reduce latency and complexity. This is facilitated by the new `@cloudflare/ai-utils` npm package, which includes the `runWithTools` function for executing LLM inference and provided function code in a single step, and `createToolsFromOpenAPISpec` for dynamically generating tools from OpenAPI specifications. The `autoTrimTools` helper function is also introduced to optimize token usage by pre-selecting relevant tools.

6/24/2024

This post details the development and implementation of Cloudflare's Bot Management machine learning model v8, which specifically targets and detects bot attacks utilizing residential proxies. It explains the challenges posed by residential proxies, the evolution of Cloudflare's bot detection models since 2019, the training methodology for v8 including data sourcing and validation, and the specific ML features used. The post highlights how v8 analyzes network and behavioral signals to identify residential proxy abuse without relying on IP blocking, thus reducing false positives.

5/22/2024

This post announces the General Availability of AI Gateway, an AI ops platform that acts as a proxy for generative AI workloads. It provides a unified interface for managing and scaling AI applications, offering features such as analytics, real-time logs, caching, rate limiting, support for multiple AI providers (including Workers AI, Groq, and Cohere), and universal endpoints with request fallbacks. The post details the technical implementation of these features and outlines future development plans.

4/18/2024

This post announces the availability of Meta's Llama 3 8B instruction-tuned model on Cloudflare Workers AI. It details the technical aspects of Llama 3's architecture, including its increased training tokens, larger context windows (with current support for 2.8k and planned support for 8k), and its efficient tiktoken-based tokenizer with a 128k vocabulary. It also highlights the use of grouped-query attention (GQA) for inference efficiency and byte-pair encoding (BPE) for tokenization. The post provides a code example demonstrating how to run inference with Llama 3 on Workers AI and mentions upcoming support for other Meta AI offerings like Llama Guard 2, Code Shield, and CyberSec Eval 2.

4/2/2024

This post announces the General Availability (GA) of Workers AI, a serverless GPU inference platform. Key improvements include enhanced load balancing for increased reliability and performance, routing requests to GPUs in over 150 cities globally, and significantly increased rate limits. Pricing has been optimized, making popular models like Llama 2 and Mistral 7B considerably cheaper. New dashboard and playground tools are introduced for better analytics and model testing. The partnership with Hugging Face is expanded, offering more curated models and a direct integration button. A significant new capability is Bring Your Own LoRA (BYO LoRA) support for fine-tuned inference, allowing users to apply LoRA adapters to base models at runtime. Additionally, Cloudflare Workers now support Python, enabling developers to write Workers in Python using frameworks like LangChain and Pyodide.

4/2/2024

This post introduces the open beta of fine-tuned inference with LoRAs on Workers AI, supporting Mistral, Gemma, and Llama 2 models. It details the technical concepts of fine-tuning and LoRA, explaining how LoRA adapters reduce trainable parameters and GPU memory requirements. The post also dives into the implementation details of multi-tenant LoRA serving on Cloudflare's platform, highlighting the use of the Punica CUDA kernel and global cache optimizations to manage GPU resources efficiently and achieve low inference latency.

3/14/2024

This post details the discovery and mitigation of a token-length side-channel attack affecting AI products that use streaming responses. Cloudflare collaborated with researchers to patch a vulnerability in Workers AI and AI Gateway by adding a 'p' property with a variable random length string to JSON objects in streaming responses, obscuring token lengths and preventing attackers from inferring plaintext information. This mitigation is automatically applied to all users of Workers AI and AI Gateway.

3/4/2024

This post announces the development of 'Firewall for AI,' a specialized Web Application Firewall (WAF) designed to protect Large Language Models (LLMs) from emerging threats like prompt injection and sensitive data disclosure. It highlights the unique attack surfaces of LLMs compared to traditional applications and details how Firewall for AI will integrate existing WAF tools (Rate Limiting, Sensitive Data Detection) with new prompt analysis capabilities to secure LLM deployments across internal, public, and product models.

3/4/2024

This post details how Cloudflare is leveraging its existing AI and machine learning capabilities to combat AI-enhanced phishing and Business Email Compromise (BEC) attacks. It explains how Large Language Models (LLMs) can be used by attackers to create more convincing phishing emails, particularly for BEC scams, by improving grammar, flow, and personalization. The post highlights that despite these advancements, LLM-generated attacks still rely on fundamental attack vectors like malicious links and downloads, and are limited by factors like opportunity and access to data. Cloudflare's existing email security models, trained on billions of messages, are already effective at detecting these sophisticated threats by analyzing multiple signals beyond just the text, including sender reputation, correspondence patterns, metadata, domain authenticity, and communication graphs. The post emphasizes that while generative AI introduces new challenges, Cloudflare's robust, multi-layered security approach, powered by AI, is well-positioned to defend against these evolving threats.

3/4/2024

Introduces an AI assistant for Cloudflare's Security Analytics dashboard. This feature leverages the Workers AI platform and off-the-shelf LLMs to convert natural language queries into GraphQL filters for security event data. The system generates time series charts based on these queries, enabling users to more quickly discover anomalies and potential security attacks without manual filter configuration. It emphasizes privacy by keeping customer data within Cloudflare's network and not using it for model training.

3/4/2024

This post introduces Cloudflare's 'Defensive AI' framework, detailing how AI is being integrated into security solutions to combat AI-powered threats. Key developments include: 1. API Anomaly Detection for API Gateway, an upcoming ML-powered feature that learns application business logic by analyzing client API request sequences to identify deviations and stop attacks. 2. Enhancements to the Web Application Firewall (WAF) using ML models (like WAF Attack Score) that learn from attack traffic to detect new and evolving vulnerabilities without manual rule creation. 3. AI models within Cloudflare Email Security (Honeycomb and Labyrinth) to detect sophisticated phishing attacks, analyzing sender reputation and spoofing attempts. 4. AI-driven user risk scoring for Cloudflare Zero Trust, which analyzes real-time user behavior to identify anomalies and tailor security posture.

3/3/2024

This post marks a significant acceleration in Cloudflare's AI integration, specifically highlighting the development of new products and features designed to safeguard AI models and applications, prevent data leakage from AI usage, and leverage AI for enhanced security against AI-enhanced phishing. It also emphasizes Cloudflare's philosophy of using AI to bolster defenses against sophisticated attacks and improve customer interaction with analytics.

2/29/2024

Introduced a new pricing model for Zaraz based on 'Zaraz Events' (pageviews, track events, etc.), offering 1,000,000 free events per month per account, with additional events priced at $5 USD per million. All Zaraz features are now available on all plans (Free to Enterprise). Migrated all tools to Managed Components for granular permission control and privacy enhancements. Introduced the Zaraz Context Enricher for custom data manipulation in Cloudflare Workers.

2/28/2024

This post announces the addition of 17 new models to the Workers AI catalog, expanding its capabilities for text generation, summarization, and image generation/manipulation. It highlights new LLMs (e.g., DeepSeek Math, Qwen 1.5 series, Falcon-7b-instruct), SLMs (e.g., TinyLlama, Phi-2), summarization models (BART-large-CNN), and image generation models (Stable Diffusion XL Lightning, Stable Diffusion v1.5 inpainting/img2img). It also details new use cases like math problem solving, SQL code generation, and image captioning. The post emphasizes optimizations using ONNX and TensorRT for Stable Diffusion pipelines and introduces a revamped developer documentation for easier model discovery.

2/16/2024

This post details Cloudflare's approach to monitoring machine learning models used for bot detection. It explains the importance of monitoring for maintaining model performance in production, especially against evolving threats. The post outlines the process of using ground-truth datasets, specializations for granular analysis, and the integration of monitoring into the Endeavor ML platform. It also describes the technical implementation of rendering monitoring queries, computing performance metrics, storing metrics, and running tasks on GPUs.

2/6/2024

This post announces the addition of eight new models to the Workers AI catalog, including text generation models like Llama-2-13b-chat-awq, Zephyr-7b-beta-awq, Mistral-7b-instruct-v0.1-awq, OpenHermes-2.5-Mistral-7b-awq, Neural-chat-7b-v3-1-awq, and Llamaguard-7b-awq, as well as code generation models like Deepseek-coder-6.7b-base-awq and Deepseek-coder-6.7b-instruct-awq. It highlights the performance advantages of Deepseek-coder and OpenHermes-2.5-Mistral-7b due to their training data and fine-tuning. The post also introduces Activation-aware Weight Quantization (AWQ) as a technique for improving memory efficiency in LLMs. A new use case, prompt and response classification with LlamaGuard, is detailed with a code example demonstrating how to use it to detect unsafe content.

1/31/2024

This post details five new integrations between LangChain and Cloudflare's developer platform: Workers AI Chat Models, Workers AI Instruct Models, Workers AI Text Embeddings Models, Vectorize Vector Store, and Cloudflare D1-Backed Chat Memory. It also highlights a Nuxt + Vue frontend template for building retrieval augmented generation (RAG) chatbots using these integrations.

12/12/2023

This post significantly contributes to the AI feature thread by introducing a new category for 'Generative AI' in Cloudflare's analysis of internet services. It details the rapid rise of Generative AI platforms like OpenAI (ChatGPT, GPT-4) and Google's Bard, tracking their popularity and evolution throughout 2023 based on DNS query data. It also mentions other AI services like character.ai, QuillBot, Hugging Face, and Poe AI, providing insights into their relative performance and growth. This post showcases Cloudflare's capability to analyze and report on the impact of emerging AI technologies on internet traffic and user behavior.

12/7/2023

This post details Cloudflare's internal ML Ops platform, focusing on the training side of the ML equation. It covers the use of JupyterHub on Kubernetes for scalable notebook environments, GitOps with ArgoCD for continuous delivery and infrastructure management, curated model templates for training, batch inference, stream inference, and explainability, and various orchestration tools (Airflow, Argo Workflows, Kubeflow Pipelines, Temporal) based on team and use case preferences. It also touches on hardware considerations for GPUs and CPUs.

11/23/2023

This post announces the availability of Stable Diffusion and Code Llama models within Workers AI, accessible via API and within Workers. It provides code examples for generating images with Stable Diffusion and code with Code Llama, and highlights the expansion of Workers AI inference to 100 cities globally.

11/21/2023

This post announces the addition of the Mistral-7B-v0.1-instruct model to Workers AI, providing code examples for both REST API and Worker script usage. It delves into the technical details of attention mechanisms in large language models, specifically explaining Scaled Dot-Product Attention, Multi-Head Attention, Multi-Query Attention, and Grouped-Query Attention (used by Mistral 7B), including diagrams and performance implications. It highlights Mistral 7B's advantages in speed and performance due to grouped-query attention and sliding window attention.

11/14/2023

Introduced streaming responses for all LLMs on Workers AI using Server-Sent Events (SSE), enabling token-by-token output for improved user experience. Added a full-precision (fp16) Llama-2 model variant and increased context and sequence lengths for existing models to support longer inputs and outputs.

9/29/2023

This post details Cloudflare's participation in the IETF's Privacy Preserving Measurement working group and their implementation of the Distributed Aggregation Protocol (DAP) in their open-source aggregator server, Daphne. It explains the core principles of DAP, including data minimization and the use of secret sharing and multi-party computation to aggregate data without exposing individual measurements. The post also discusses challenges like input validation and non-linear aggregation, and how zero-knowledge proofs and Prio are used to address them.

9/29/2023

This post introduces new bot categories within Cloudflare's WAF and Rules Engine, specifically enabling users to easily allow or block AI crawlers. It highlights the existing protection against malicious crawlers and expands on the Verified Bot directory by adding AI Crawler as a distinct category. The post also outlines criteria for 'respectful AI bots' and discusses the development of industry-wide protocols for AI crawler management, demonstrating a proactive stance on integrating AI responsibly.

9/28/2023

This post introduces three new integrations to the Cloudflare Integrations Marketplace for Workers: Sentry for application monitoring (errors, exceptions, logs), Turso for edge-hosted distributed databases (libSQL/SQLite), and Momento for serverless caching. The Sentry integration uses a Tail Worker to capture and send data, with future plans for source map uploads and deployment correlation. The Turso integration leverages its global replicas and Smart Placement for read-heavy workloads. The Momento integration uses an OAuth2 flow to add API keys as secrets for SDK usage.

9/27/2023

This post announces the launch of Workers AI for serverless GPU inference on Cloudflare's global network, Vectorize as a vector database for storing embeddings, and AI Gateway for caching, rate limiting, and observing AI deployments. It also highlights partnerships with NVIDIA, Microsoft, Hugging Face, Databricks, and Meta, and discusses the strategic placement of inference workloads on the network edge. The post details the use of ONNX runtime for seamless model execution across devices, edge, and cloud, and the integration of Llama 2 and Hugging Face models.

9/27/2023

Introduced Workers AI, a serverless platform for running AI inference on Cloudflare's global network of GPUs. This enables developers to run popular open-source AI models (e.g., Llama 2, Whisper, ResNet-50) with minimal code via a REST API or Cloudflare Workers/Pages. Key features include privacy by default, pay-as-you-go pricing, and integration with Vectorize for vector database functionality. The service plans to expand its model catalog through a partnership with Hugging Face and increase GPU deployment globally.

9/27/2023

This post details the practical implementation of running the Llama 2 LLM on Cloudflare Workers AI. It provides a 14-line JavaScript code example for deploying the model and demonstrates its usage with various prompts, including generating poems and contrasting concepts. The post highlights the ease of use and rapid deployment of AI models on Cloudflare's global network.

9/27/2023

Introduces AI Gateway, a new product that acts as a proxy between applications and AI APIs (e.g., OpenAI, Hugging Face, Replicate). AI Gateway provides caching, rate limiting, request retries, and analytics for AI applications. It offers dedicated endpoints for specific providers and a universal endpoint for flexible routing and fallbacks. The post details how to integrate AI Gateway with a single line of code and showcases its analytics capabilities for monitoring requests, users, costs, and duration. It also highlights the integration with Cloudflare Workers AI and Vectorize for building full-stack AI applications.

9/27/2023

This post announces a partnership with Hugging Face to make AI model deployment easier and more affordable. Key contributions include bringing serverless GPU models to Hugging Face (eliminating infrastructure management and pay-for-unused capacity), integrating popular Hugging Face optimized models into Cloudflare's model catalog, and introducing Cloudflare integrations into Hugging Face's Inference solutions. This expands Cloudflare's AI offerings by providing direct access to a vast library of AI models and simplifying their deployment on Cloudflare's platform.

9/27/2023

Introduced Vectorize, a new vector database service in open beta, designed for building AI-powered applications on Cloudflare's global network. Vectorize integrates with Workers AI and custom embeddings to enable semantic search, classification, recommendation, and anomaly detection. The post details the necessity of vector databases for providing context to ML models and LLMs, contrasting the inefficient approach of passing large datasets with the efficient workflow of using a vector database. It also provides a code example demonstrating how to insert and query vectors using the Vectorize API within Cloudflare Workers.

9/27/2023

This post details how several AI companies (Azule.ai, 42able.ai, Layercode, Eclipse AI, Embley, Chainfuse) are building and deploying AI applications using Cloudflare's developer platform. It highlights the use of Durable Objects, WebSockets, Queues, R2 for vector storage, Workers KV, D1, Workers AI (Constellation), Workers for Platforms, and Pages. Specific examples include Azule.ai using R2 for on-disk vector search, Layercode using Workers for voice AI pipelines and Durable Objects for hosted agents, and Eclipse AI leveraging R2 for object storage and Workers for backend APIs. The post emphasizes the benefits of Cloudflare's platform for AI development, such as reduced costs, faster development cycles, and improved performance.

9/27/2023

This post announces Cloudflare's rebranding as a 'connectivity cloud' and details a major expansion into AI, specifically focusing on providing GPU resources for AI inference across their global network. It contrasts this approach with centralized public clouds and on-device processing, highlighting the benefits of their distributed model for speed, privacy, and cost. The post also references past birthday announcements that have significantly impacted the internet, positioning this AI expansion as a similarly transformative event.

7/19/2023

This post announces the general availability of Cloudflare Zaraz, a solution for server-side data collection and management of third-party tools. It details the product's evolution from beta, its integration into the Cloudflare Developer Platform, and its new pricing structure across Free, Paid (Workers), and Enterprise plans. The pricing is based on 'Zaraz Loads' (page loads or Pageview trigger activations) and feature sets. The post highlights Zaraz's technical underpinnings on Cloudflare Workers, enabling features like Worker Variables and server-side data collection to improve website performance, security, and privacy. It also provides guidance for existing beta users on how to transition to the new pricing model.

6/27/2023

Introduced JSONata integration within Cloudflare Zaraz, allowing users to perform data transformation and manipulation on payloads sent to third-party tools. This provides a middle-ground solution between simple field mapping and full Cloudflare Worker integration for data processing.

6/22/2023

Introduced three new features for Constellation: increased model size limit from 10MB to 50MB, added tensor caching for input tensors to reduce network and parsing overhead, and integrated the XGBoost runtime alongside the existing ONNX runtime. This post also elaborates on the concept of globally distributed AI and its use cases, including personalized web experiences and making devices smarter.

6/19/2023

This post details the evolution of Cloudflare's machine learning feature serving infrastructure. It describes the transition from a previous system using Gagarin and Unix Domain Sockets (UDS) to a new approach leveraging memory-mapped files for inter-process communication (IPC). The new system addresses issues of high tail latency, suboptimal resource utilization, and scalability constraints by employing wait-free synchronization and efficient data structures to serve hundreds of float features for millions of dimension keys with ultra-low latency.

6/19/2023

This post details optimizations applied to the machine-learning detections within Cloudflare's bot management module, which accounts for approximately 15% of the latency added by bot detection. By porting from Lua to Rust, eliminating memory allocations, and optimizing parsers, the P50 latency of the bot management module was reduced by 79μs (a 20% reduction). Specific techniques include using fixed-size buffers on the stack, choosing algorithms that operate in-place, and leveraging testing tools like `dhat` to ensure zero allocations. Optimizations were also applied to the CatBoost ML library, including optimizing for single-document evaluation and reusing buffers for categorical features, leading to up to 15% performance gains in production models.

5/18/2023

Introduces Cloudflare Secrets Store, a new secrets management offering. This expands on previous capabilities like environment variables and secrets for Cloudflare Workers by providing account-level secrets that can be referenced across multiple Workers and other Cloudflare services like WAF. It emphasizes security features such as scoped permissions, audit logs, and on-demand encryption key rotation. The post highlights the evolution from narrowly scoped Worker secrets to a centralized solution for managing sensitive data across the Cloudflare ecosystem.

5/18/2023

This post details the integration of LangChainJS with Cloudflare Workers, enabling developers to build AI-powered applications at the edge. It provides a step-by-step guide on setting up a Cloudflare Worker, installing LangChainJS, and using document loaders (specifically CheerioWebBaseLoader) to fetch and process web content. The post also introduces the concepts of text splitting for managing LLM input size and the role of embeddings and vector stores in understanding text relevance for LLM interactions.

5/15/2023

Introduced Constellation, a new service that allows developers to run pre-trained machine learning models and inference tasks on Cloudflare's network using Cloudflare Workers. Constellation supports various ML models (e.g., SqueezeNet, T5) and integrates with other Cloudflare services like Email Routing. The post provides examples of image classification and sentiment analysis using Constellation, detailing the development workflow with Wrangler and the Constellation client API.

5/15/2023

This post details how Cloudflare One's Zero Trust security tools can be applied to manage the risks associated with using AI services. It introduces features for measuring AI usage (observing SaaS application traffic), controlling API access to sensitive data using service tokens and Cloudflare Tunnel, restricting data uploads with Data Loss Prevention (DLP) policies, and scanning SaaS applications for misconfigurations using Cloud Access Security Broker (CASB).

5/15/2023

Introduces Cursor, an experimental AI assistant for Cloudflare's Developer Platform, trained to answer questions about Cloudflare's documentation. Cursor is built using Workers, Durable Objects, OpenAI, and the Cloudflare developer docs, employing a 'Search-Ask' method. This represents a step towards AI-driven developer experience by providing a conversational interface to documentation.

5/14/2023

This post details the upcoming integration of AI assistants into the Cloudflare Workers platform to enhance developer productivity and broaden access to the Cloudflare Developer Platform. It highlights the importance of AI capabilities within developer platforms, drawing parallels to essential features like floating-point arithmetic and data handling. The post also discusses the need for data fluidity in AI systems and how Cloudflare's R2 storage addresses this by enabling data movement without transfer fees. It frames AI as a transformative technology akin to electricity, predicting its pervasive impact on programming and human interaction with software, and notes the potential for AI to improve human performance, citing the example of Go players improving after playing against AI.

10/27/2022

This post details the formal, computer-aided security analysis of the Oblivious HTTP (OHTTP) protocol, a new privacy-preserving protocol for decoupling 'who' from 'what' was sent. It describes the protocol's design, including its use of HPKE and Binary HTTP, and presents a simplified security model and goals. The analysis uses the Tamarin prover to model the protocol's behavior and prove its privacy properties against an attacker who can compromise either the relay or the gateway, but not both.

9/22/2022

This post introduces Cloudflare Zaraz's new Managed Components and Data Loss Prevention (DLP) features. Managed Components offer a more secure and privacy-aware way to load third-party tools by running them in an isolated environment within a Component Manager. This allows for granular control over permissions and the data exposed to these components. The DLP features further enhance privacy by enabling the scanning, reporting, masking, and removal of PII from information shared with third-party tools, based on content rather than just field type. This provides a robust mechanism for fencing information before it is accessed by components.

2/21/2022

This post introduces the concept of quantum computing as a new frontier in computation, drawing parallels to the paradigm shift brought by quantum mechanics in physics. It explains qubits, superposition, and the potential for quantum computers to solve specific problems exponentially faster than classical computers, highlighting both the 'solace' of new computational capabilities and the 'spectre' of threats to current cryptography. It also discusses the challenges in building quantum computers and the ongoing research and development in the field.

1/26/2022

This post significantly expands the understanding of Cloudflare's evolving capabilities by highlighting the increasing importance of APIs in internet traffic. It details how Cloudflare is leveraging its global network data to analyze API trends, identify key industries driving this growth (like Banking and Financial Services), and understand API characteristics (methods, payload encoding). Crucially, it introduces Cloudflare's API Shield product suite (API Discovery, Schema Validation, mTLS, API Abuse Detection) as a direct response to the challenges posed by API traffic, demonstrating a proactive approach to securing this critical communication layer. The post also notes the adoption of newer protocols like TLS v1.3, QUIC, and gRPC for APIs, indicating an evolution in the underlying technologies Cloudflare supports and secures.

10/1/2021

This post introduces an open-source, full-stack application demonstrating how to mint and manage NFTs using Cloudflare Workers and Cloudflare Pages. It details the architecture, including a smart contract deployed on Ethereum's Rinkeby network, a web interface for claiming NFTs, and Cloudflare Workers for generating JSON metadata. It also highlights the use of Infura for pinning and Cloudflare IPFS Gateway for serving NFT images, showcasing how Cloudflare's developer platform can scale Web3 applications.

4/13/2021

This post announces the integration of NVIDIA GPUs and TensorFlow into the Cloudflare Workers platform, enabling AI-based applications to run at the edge. It details the partnership with NVIDIA, the benefits of edge AI for performance and security, and showcases a demo application 'nataornot.com' that uses a TensorFlow model on an NVIDIA A100 GPU.

10/9/2020

This post details an exploration of deploying AI services using WebAssembly (WASM) on Cloudflare Workers, in collaboration with Latent AI. It highlights the benefits of edge AI for scalability and privacy, and introduces Latent AI's technology for compressing AI models (e.g., MobileNetV2) using quantization to 8-bit precision, achieving significant size reduction and performance improvements (e.g., 8x faster inference compared to Tensorflow.js). The post presents experimental results on memory and inference times for different model sizes and bit precisions, demonstrating the feasibility of running optimized AI models at the edge with low latency.

6/15/2020

This post details three data science and machine learning techniques applied to improve customer support and identify attack traffic: 1. Language Classification: Developed a model that combines off-the-shelf classification with browser Accept-Language headers and country-associated languages to improve accuracy, especially for short messages. 2. Fuzzy String Matching: Evaluated and adopted the Cosine similarity algorithm over Levenshtein distance for its O(k+l) computational complexity and better performance in classifying support tickets. Introduced optimizations like a new tokenization approach and FastText for candidate value determination. 3. Threat Identification: Utilized the relationship between 499 error codes and 5xx HTTP status codes to differentiate DDoS attacks from origin issues. Further, employed 3D clustering of anonymized HTTP header data (unique zones under attack, success rate, variety ratio) to fingerprint credential stuffing attacks and identify patterns not visible on a request-to-request basis.

2/20/2019

This post details Cloudflare's practical approach to productizing machine learning, outlining a three-stage lifecycle: Business Intelligence, Standalone Machine Learning, and Machine Learning Productization. It uses dental insurance as an analogy and then applies these concepts to Cloudflare's bot management. Specifically, it describes how Cloudflare uses its vast network data (660 billion requests/day) to train ML models for bot detection, achieving a 99.68% true positive rate for credential stuffing attacks on WordPress login pages. It also discusses training a model to predict which companies will be attacked next based on firmographic data, finding that smaller companies are more vulnerable.