Check what your AI agents do before they do it, with Semantic governance policies
Blog post from Google Cloud
Semantic governance policies, available in Preview on the Gemini platform, add an inline check to agent tool calls before they execute, comparing each proposed action with both the user’s request and enterprise rules written in plain language. Enforced through Agent Gateway, an LLM judge evaluates the call in the context of the conversation, agent instructions, tool descriptions, and policies; approved calls proceed, while denied calls are replaced with a configurable message, and decisions and rationales are recorded in Cloud Logging. For example, a policy can block an agent from sharing unreleased financial results with investors or personal email accounts even if the user requests it, while an intent check can stop an unrequested action prompted by hidden text in a document. The feature is designed to complement, not replace, access controls and other safeguards; because its judgments are probabilistic, Google recommends testing policies in dry-run mode before enforcement. Text-only responses and streamed model responses are not evaluated.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.