Should Agents Govern Themselves? AAV Adds an External Action-Verifier Layer
CarlosMarreroAAV · reddit · 2026-09-04
The author argues that as AI agents gain access to real systems (APIs, databases, email, MCP tools, infrastructure), a prompt telling the agent not to do something is not a security boundary. Actions like prod.database.delete() shouldn't be authorized by the agent itself. He proposes an independent control layer between intent and execution answering: who's requesting, is this agent allowed this tool, what policy applies, how risky, does a human need to approve, what exactly executed, and can audit history be verified. His product AAV (Agent Action Verifier) implements: request → identity+policy+risk evaluation → ALLOW/DENY/REQUIREAPPROVAL → authorized execution → verifiable receipt. It's running in production with 15 days free access for developers willing to stress-test it. The open question: should governance live inside the agent/framework, or should an agent never be trusted to govern its own actions?
More from coding & agent
- Dev: Codex optimized for hours against the wrong label column, 'deeply uncurious' — ivan_bezdomny · 2026-09-04
- No-code dev shares full AI pipeline for cozy game: Opus, Gemini, Meshy, Claude, Unity — Gambo7592 · 2026-09-04
- 43 Projects From Singapore's ChatGPT Sites Hackathon Open for Voting — gabrielchua · 2026-09-04
- img2threejs explained: turning one photo into editable Three.js code with 80k-180k tokens — maier_ak · 2026-09-04
- Dev Adds Voice Chat to 'Ghost' AI Assistant So It Can Check His Agent Orchestrator — BLUECOW009 · 2026-09-04
- Practitioner: Tool-Call Trajectories Are Higher-Signal Than CoT for Agent Monitoring — joshua_saxe · 2026-09-04