AWS Turns Natural-Language Policies into Dogwood Agent Governance Rules in Bedrock AgentCore

TechCrunch AI · rss · 2026-08-21

An AWS ML Blog post covers the expanded Policy Authoring capability in Amazon Bedrock AgentCore: an AI-driven tool that converts natural-language policy documents into formally correct Dogwood—the open source governance language—enforced in real time by the Dogwood monitor built into the AgentCore Gateway.

The new capabilities cover three kinds of policies:

Using a retail-bank servicing agent (verifyidentity, issuerefund, initiatetransfer, etc.), the post walks through full translation examples. Dogwood is default-deny with forbid overriding permit; the tool schema fed to authoring is generated from the agent's MCP tool manifest so generated policies reference the names the agent actually calls. Note: Policy Authoring is a translator, not a summarizer—strip rationale and background from input documents down to the rules themselves first.

Original post →

More from coding & agent

coding & agent channel →