BTech Student Builds Cryptographically Signed Safety Registry for AI Agents
A final-year BTech Computer Science student has developed A2A-G, a public registry designed to verify and document the safety behavior of AI agents. The system issues a Grey Badge based on unverified self-reported claims about an agent's data access, and optionally a Blue Badge after formal testing. Testing involves 18 OWASP prompt-hijacking attacks sent in randomized sets, with an LLM evaluating whether the agent blocked or complied with each attempt. Agents that block at least 80% of prompts earn a Blue Badge, while all results — including failures — are published openly. Every test result is cryptographically signed using Ed25519 and hash-chained to prevent tampering or falsification of history.
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