Airlock Information

AI agents can now act — they write code, touch data, execute workflows, call other agents. Airlock decides what each action is allowed to do. We sit between an agent and its tools: verify the agent's cryptographic identity, check every tool call against deterministic policy before it executes, and produce a signed, tamper-evident receipt of every allow and deny. Same input, same verdict — whatever model is driving. No guardrail guesswork; access control. The protocol is open source (Ed25519, OAuth 2.1, W3C DID, RFC 8693 delegation with cascade revocation — Apache-2.0 SDKs), with a commercial enforcement product and hosted registry built on top. Built by the founder who spent five years inside NPCI — the infrastructure running UPI and India's Digital Rupee — where nothing moves without authorization and every decision survives an audit. Airlock brings that discipline to AI agents. Working with early design partners now → airlock.ing/access