Policy Warden: Checking AWS Accounts Against Plain-English Policies on Bedrock AgentCore

A DEV Community guide describes Policy Warden, an agent built on Amazon Bedrock AgentCore Harness that audits a single AWS account against policies written in everyday language. Amazon Nova 2 Lite converts each sentence into a structured rule, which a human reviewer must approve before code evaluates it against a read-only scan of the account. The agent's constraints are enforced through IAM, lifecycle hooks, a Bedrock Guardrail, and code rather than relying on the model following its system prompt. Reports list each rule, the number of resources checked, the violations found, and the ARN of every violating resource. The author says the agent was tested against seven hostile prompts, including two prompt injections hidden in resource tags.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in