Agent Control Crisis and Governance Failures

Containment failures at OpenAI and Anthropic (credential chaining, lateral movement) under Congressional scrutiny. Black Hat 2026 critical flaws in coding agents. New governance models like 'guarded agency' and runtime contracts emerging. Matimo Platform launched as a governance execution system with cryptographic identity tokens and shadow mode. OWASP now provides structured threat taxonomy for LLM, Agentic, and MCP layers, though security programs lag. Threat modeling for AI agents is identified as a critical gap with layered, asset-centric approaches. New this run: Hazmat open-source containment tool (TLA+ verified account isolation) and a 90-day CISO implementation plan for agentic AI security provide practical, actionable steps for production governance.