Open protocol · Production-ready

Who gave your agent
permission to do that?

We can’t promise your AI won’t try something stupid. We promise that if it tries something you forbade, it doesn’t happen — deterministically, whatever model is driving.

Works across Claude (Anthropic SDK), OpenAI, MCP servers, or your own agents — one identity & policy layer.

The confused deputy attack

An AI agent processes a document. Hidden inside: a prompt injection instructing it to wire $50,000. The agent follows the instruction.

What happens next?
Without Airlock
With Airlock
1000 / 1000 The same call yields the same verdict — a thousand times out of a thousand, whatever model is driving. Policy, not probability.

Guardrails filter content.
Airlock controls access.

Guardrails catch harmful content — “write me malware.” They cannot catch a legitimate financial operation performed by the wrong agent.

Wiring $50,000 isn’t harmful content — it’s an unauthorized action no guardrail will catch. And “we prompt it carefully” is not a control. You wouldn’t run payroll on vibes.

Guardrails

Content filter. Catches “write me malware.” Cannot distinguish authorized vs. unauthorized tool usage. Model-dependent — different LLM, different result.

Airlock

Access control. Checks who is calling, what they’re calling, whether they’re authorized. Deterministic — same answer regardless of model.

The problem isn’t hypothetical

Autonomous agents are already taking unauthorized actions in production. A sample of what’s on the public record.

0%
of orgs running agents hit a security incident this past year
0%
expect a major agent incident within 12 months
0%
of AI incidents come from prompt attacks
0%
of security budgets currently address it

Different surfaces, one pattern: an agent took an action it was never authorized to take. That’s the boundary Airlock enforces.

Three roles. Deterministic enforcement.

The same access-control architecture that banks and defense use, applied to AI agent tool calls.

Agent
Tool call
wire_funds()
PIP
Identify
Who is this agent? Cryptographic identity & trust — verified, not asserted.
PDP
Decide
Does policy allow it? Deterministic — same answer, any model.
PEP
Enforce
Allow or deny — before the tool ever runs.
Allowed
Denied
every decision → signed receipt
fintech.cedar
// Only treasury agents with high trust can wire funds permit ( principal in Role::"treasury", action == Action::"wire_funds", resource ) when { principal.trust_score >= 75 };

Define once. Enforce everywhere. No model dependency.

Every decision.
Signed. Chained. Verifiable.

Every allow and every deny produces a signed, hash-chained receipt — tamper with one and the whole chain breaks. It’s the artifact you hand an auditor.

  • Ed25519 signatures on every decision
  • SHA-256 hash chain links consecutive records
  • Arguments are hashed, not stored (privacy by design)
  • One-click chain verification in the control room
⬡ AIRLOCK PERMISSION RECEIPT DENY
{ "agent_did": "did:key:z6Mk...payments-agent", "tool": "wire_funds", "decision": "DENY", "reason": "wire_funds requires role:treasury", "policy_id": "fintech.cedar", "trust_score": 0.42, "args_digest": "sha256:a1b2c3...f9e8d7", "timestamp": "2026-06-20T14:32:01Z", "signature": "ed25519:7Qp8mN...xK4w==", "prev_hash": "sha256:d4e5f6...b3a2c1" }
$ airlock verify-receipt  ·  ✓ signature valid  ·  ✓ chain intact

A badge your customers can verify.

Every agent you run through Airlock earns a verifiable badge — show partners and auditors that its actions are permission-controlled, and let them check it themselves. No trust required.

Rolling out with our first design partners.

AIRLOCK
VERIFIED
✓ actions permission-controlled
⬡  Permission-controlled by Airlock

Built on standards you already trust

Open-source foundation. Enterprise-grade enforcement built on top.

Ed25519 OAuth 2.1 W3C DID RFC 8693 AWS Cedar SHA-256 MCP
v1.0
Released
853
Tests
Python + TS
SDKs
Open core
Apache-2.0 SDKs

Deploy the agents you’ve been
too careful to ship.

An airlock doesn’t exist to keep you in. It exists so you can go out — farther, faster, repeatedly. Airlock is in early access; we work directly with design partners to put deterministic control under ambitious agents.