Check what your AI agents do before they do it, with Semantic governance policies
Check what your AI agents do before they do it, with Semantic governance policies
Google Cloud's Semantic governance policies, in Preview in the Gemini platform, check each tool call an agent proposes before it runs against plain-language rules and against what the user originally asked for, enforced through a policy enforcement point such as Agent Gateway. A call must pass both checks; denied calls are replaced with a configurable denial message, and verdicts and the judge's rationale go to Cloud Logging. The post notes the LLM judge is probabilistic, recommends dry run mode before enforcement, and says streamed model responses are not evaluated.
Useful for teams deploying agents that can send email or share files, since it shows a concrete pre-execution guardrail against data leaks and indirect prompt injection.
Source: Google Developers Blog · developers.googleblog.com