TechPrufer
Get started
Google
Google logoGoogle · Feature trail

Campaign Anomaly Detection

2 posts · Jul 2022 – Sep 2026

What this trail covers

Agent Anomaly Detection is a reasoning-based oversight and audit layer for autonomous agents deployed on the Gemini Enterprise Agent Platform. It examines what an agent actually does using its reasoning traces, tool calls, and execution flow across a session. It reads the logs and OpenTelemetry traces your agents already emit, evaluates that activity to decide whether an agent is operating outside its intended boundaries, and flags behavioral anomalies, suspicious intent, and policy violations. ShieldGemma 2, built on Gemma 3, is a 4 billion (4B) parameter model that checks the safety of your synthetic and natural images against key categories to help you build robust datasets and models. With this addition to the Gemma family of models, researchers and developers can now easily minimize the risk of harmful content in their models across key areas of harm: Sexually explicit content, Dangerous content, Violence. We recommend using ShieldGemma 2 as an input filter to vision language models, or as an output filter of image generation systems. ShieldGemma can be used on both synthetic and natural images.

All Google trails
  1. Agent Anomaly Detection, now in Private Preview on the Gemini Enterprise Agent Platform- Google Developers BlogIntroduces Agent Anomaly Detection, a new reasoning-based oversight and audit layer for autonomous agents on the Gemini Enterprise Agent Platform. It analyzes agent behavior through reasoning traces, tool calls, and execution flow to detect anomalies, suspicious intent, and policy violations. Key features include no added runtime latency, actionable findings mapped to OWASP Agentic Top 10 risks (tool misuse, identity/privilege abuse, cascading failures, rogue agents, resource exhaustion, token usage escalation), and upcoming custom business logic detection. The detection pipeline uses layered analysis: a lightweight first pass for statistical anomalies, followed by an LLM-based reasoning layer for deep examination. It provides findings with severity, probability, rationale, and recommended fixes, publishing to Security Command Center and exposing an API for programmatic action. It is in Private Preview for ADK 1.2+.
  2. Start building with Gemini 2.5 Flash- Google Developers BlogIntroduces Gemini 2.5 Flash, a new model with hybrid reasoning capabilities and controllable thinking budgets. Developers can now set a 'thinking budget' (0 to 24576 tokens) to fine-tune the trade-off between reasoning quality, cost, and latency. The model can perform a 'thinking' process before responding, improving accuracy on complex tasks. This offers a cost-efficient thinking model with comparable metrics to other leading models at a fraction of the cost and size.
  3. Safer and Multimodal: Responsible AI with Gemma- Google Developers BlogIntroduces ShieldGemma 2, a multimodal safety classifier built on Gemma 3, designed to detect harmful content in synthetic and natural images. It extends the capabilities of ShieldGemma to image safety, addressing challenges in multimodal models. The post details its performance benchmarks, flexibility for fine-tuning, versatility with Gemma 3 supporting frameworks, and collaborative nature. It also mentions plans for smaller sizes, more harm areas, and alignment with multimodal ML Commons taxonomy.
  4. Campaign Anomaly Detector- Google Developers BlogThis post introduces the Campaign Anomaly Detector, a system built to identify anomalies in campaign performance. It details the use of a Prophet model for time-series forecasting and the implementation of anomaly detection logic based on forecast deviations. The system leverages BigQuery for data storage and analysis, and Cloud Functions for triggering anomaly detection jobs. It also describes the integration with internal alerting systems to notify stakeholders of detected anomalies.