Authenticated Workflows: A Systems Approach to Protecting Agentic AI
This could finally make enterprise AI agents safe enough for real use...
Deep Dive
Researchers have introduced 'authenticated workflows,' a new cryptographic system claiming to provide deterministic security for agentic AI. It protects four boundaries—prompts, tools, data, and context—with cryptographic proofs and a new policy language called MAPL. The system integrates with nine major AI frameworks (OpenAI, Claude, LangChain, etc.) without modification. In testing, it achieved 100% recall with zero false positives across 174 cases and mitigated 9 of 10 OWASP Top 10 risks.
Why It Matters
It could unlock safe, large-scale deployment of autonomous AI agents in enterprises by eliminating probabilistic security failures.