Model Armor
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 4
- Google reports Agent Anomaly Detection benchmark results
Google said its private-preview Agent Anomaly Detection flagged 87% of harmful instructions that an agent carried out across five public, independently written security benchmarks. Performance was strongest against goal hijack and tool misuse, while poisoned memory remained a clear weakness, with only 4 of 18 harmful memory cases detected. The first-party study used Gemini 2.5 Flash and simulated applications, and produced false alarms on 2% to 16% of benign sessions depending on the tools used.
Google Developers Blog · 🔥 0 - Google Cloud FDE says small teams can now deploy production Agents
At the Silicon Valley People ACC event, Google Cloud Greater China FDE team lead Wang Shuyuan said cloud platforms are absorbing generic Agent engineering work that once required a full organization. He described a virtual IP studio delivered by one FDE engineer and one customer product manager, plus a retail Agent project handled by three engineers with less than six months of tenure. He also reported that grounding a travel-planning workflow with Google Maps raised route-image generation success from 60% to 87.5%, while a retail shopping Agent was evaluated across 88 real shopping scenarios.
品玩 实时要闻 · 🔥 0 - Anthropic-managed Claude Models now generally available on Google Cloud
Claude Blog and Google AI on DEV report that Claude Models on Google Cloud (Anthropic managed) is generally available through US and Global endpoints. According to Anthropic, the offering provides the full Claude Platform feature set, with new features and betas available on the same day they launch natively, while using existing Google Cloud IAM credentials, one invoice and applicable Google Cloud commitments. The Google-managed option remains available with Google as data processor, regional data residency and native integrations such as Model Armor and Agent Engine.
Claude Blog · 🔥 0 - Google Cloud details governed self-evolving AI agent architecture
Google Cloud introduced Project Chrysalis, a research reference architecture for governed, self-evolving AI agents that can diagnose outages, synthesize mitigation code, assess interventions, retain operational lessons and propose changes to an agent hierarchy. Built on the Gemini platform and Agent Development Kit, it uses Model Armor, Cloud KMS, Binary Authorization, Open Policy Agent and human approval gates to constrain autonomous actions. In automated Black Friday cache-stampede demo simulations, Google reported that time to first mitigation fell from 15 to 45 minutes to 1.8 seconds, HTTP 5xx error rates fell to 0.0%, and P99 latency dropped from 519.8 milliseconds to 18.2 milliseconds. The report did not provide independent production validation.
Google Developers Blog · 🔥 0
Experience and discussion from the community
Share my Model Armor experienceAsk about Model Armor
Nobody has shared their experience with Model Armor yet.