CAIN-42 CAIN StudioBROKER: one action, one use
Most AI agents run on a standing API key: one leaked prompt and the key does anything, forever. BROKER replaces it with an Action Credential: an Ed25519-signed token bound to one HTTP method, one exact path and the SHA-256 of one exact body, valid for seconds and usable once. Any service can check it offline with the published key.
#7 of the CAIN-42 Top 25 4 flaws found and fixed today fix live since 2026-11-02 15:31Z (verified on both gateway instances)
action per credential: method, exact path, exact body
default lifetime (300 s maximum)
use, shared across both gateway instances, and now across every child credential
model involvement: minting is a deterministic decision against the agent's grants
Why it matters
Non-human identity is where agent security is heading: agents need authority for the next action, not a key to everything. BROKER is that idea built and checkable: a standard compact JWS (EdDSA), a published JWKS, a downloadable verifier, and a signed log of every credential issued.
On this site
How it works
Grant, decide, mint, use once.
Delegation
Narrowing a credential for a sub-agent, safely.
Live demo
Mint, use, replay.
Evidence
Four flaws, before and after.
Verify it yourself
The offline verifier.
Path rules
Why some paths are refused.
API
Routes and the token format.
Limits
Honestly.
Use cases
Where it fits.
2026-11-02 upgrade
What we broke and fixed.
FAQ
Questions.
FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.