Governing the Ungovernable: Security and Compliance for AI Agents in Open Source Projec...
summary
Describes a working open-source governance layer called GitMesh that intercepts agent actions, compiles YAML policies to OPA, and records immutable, Sigstore-signed audit logs; shows concrete failures (an agent deleting all local containers, Replit and Cloud Code incidents) and a mitigation path: policy-as-code to OPA, human-in-the-loop gates, 53% of orgs exceed intended agent permissions; demonstrates adapters for Claude/Cursor/Codex, atomic task assignment, heartbeat scheduling, and a live control plane that enforces policies before actions execute.
AI-written summary · about
chapters (9)
- 0:00AI agents and GetMesh intro
- 3:08Risks of ungoverned AI agents
- 7:05Prompt injection causes
- 10:57Secure agent governance philosophy
- 13:08GitMesh security model pillars
- 17:25GitMesh tool integration
- 23:43GetMesh project origin
- 26:30GitMesh live demo
- 29:21Open source community engagement
related talks
51mBeyond Code Generation: Governing AI Agents Across the DevSecOps Lifecycle
23mGoverning AI Agents at the Hardware Boundary
30mGitops with Policy As Code and OPA
30mDeclarative by Default, Secure by Design: GitOps as a Control Plane for Governance
24mAI Governance for developers: A practical framework for deploying autonomous coding agents safely…
30mRedefining Access Control: Scaling Policy as Code for Humans and AI Agents
29mIntroduction to Open Policy Agent
36mHow to Safely Deploy AI Agents that Write and Manage your IaC