> On 27 October we presented MintID’s kill switch to the Security & Privacy Working Group of the Agentic AI Foundation (AAIF), which has a Kill Switch design pattern in draft. It revokes the credential’s authority; it does not stop the agent’s software. Up to 245 seconds after the chain records the kill switch, every verifier on the network refuses every copy of the credential. Blog · kill switch # Switching off an agent’s authority, and how long it takes On 27 October we presented MintID’s kill switch to the Security & Privacy Working Group of the Agentic AI Foundation (AAIF), which has a Kill Switch design pattern in draft. It revokes the credential’s authority; it does not stop the agent’s software. Up to 245 seconds after the chain records the kill switch, every verifier on the network refuses every copy of the credential. by Marc Miró i Rierola·published 2026-10-30·updated 2026-10-30·4 min read On 27 October we presented MintID’s kill switch to the Security & Privacy Working Group of the Agentic AI Foundation (AAIF), which has a Kill Switch design pattern in draft. **What gets switched off.** An approved issuer verifies a person once, with KYC. The person’s agents carry their own credentials, and each credential proves that a verified, accountable person stands behind the agent, while the person’s identity stays with the issuer and no personal data goes on the chain. Every decision checks the credential’s current status: the issuer updates it every 30 seconds, the agent answers each request with a zero-knowledge proof valid for 10 seconds, and the verifier refuses whenever it cannot prove the status. That check is where the kill switch acts. It revokes the credential’s authority; it does not stop the agent’s software. **Who can pull it.** The agent’s own runtime can pull the kill switch for its credential, and so can the person, with the agent’s revocation kit. The person can also revoke their own credential, which takes all their agents with it. The issuer can revoke anything it issued, and publish an emergency update when it is urgent. A compromised agent can block only its own runtime; it cannot stop the person or the issuer. Once a credential is revoked, verifiers refuse it, and nobody learns who pulled the switch or which credential it was. **How long it takes.** Up to 245 seconds after the chain records the kill switch, every verifier on the network refuses every copy of the credential. The bound adds up three parts: 60 seconds for the issuer to remove the credential, 180 seconds for the oldest status a verifier accepts, and 5 seconds of clock tolerance. If the issuer misses its 60 seconds, the chain refuses its status updates. Bitgrit, acting as an independent operator for the Identity & Trust Working Group in its [first external test round](/en/blog/external-test-round-1), measured one run: a copy was last accepted 27.5 seconds after the chain recorded the switch-off, and first refused at 33.8 seconds. That is an observation; the bound is what we guarantee. **How it fits the group’s pattern.** MintID covers the pattern’s credential revocation for MintID credentials, from outside the agent’s reach. It never closes again on its own: a revoked credential does not come back, and resuming means a new credential, unlinkable to the old one. Its latency is published as a bound. The rest stays with whoever deploys the agent: API keys and other tokens, halting the process, cutting its network and tools, detecting misbehaviour, and rolling back actions already carried out. An access token that a service issued from an earlier accepted presentation also runs to its own expiry. **The question we put to the group.** Should the standard ask for a maximum time for each way of revoking, and say when the clock starts? We have offered the group text for the draft along those lines: each revocation path with its bound, its start event and its acceptance window. **The deck** from the session is [attached as a PDF](/docs/MintID-AAIF-Security-Privacy-kill-switch-2026-10-27.pdf). The test harness, its records and [the public testnet](/en/developers) are open for anyone to re-run. ## The deck The deck from the session is attached as a PDF. The test harness, its records and the public testnet are open for anyone to re-run. [Deck (PDF)](/docs/MintID-AAIF-Security-Privacy-kill-switch-2026-10-27.pdf)[Developers](/en/developers) --- Source: https://mintid.net/en/blog/switching-off-an-agents-authority · Switching off an agent’s authority, and how long it takes — MintID blog