The open standard for
AI agent insurance

What is AIPS-1?

Identity tells a counterparty who an agent is. Execution-environment integrity tells the counterparty where the agent operates. Neither answers the question that determines whether a counterparty can rationally transact with an autonomous agent: if something goes wrong, who pays?

AIPS-1 addresses what we call the Illegible Insurance Problem. Cover may exist — the agent operator may have written a perfectly serviceable policy with a regulated insurer — but the cover is not legible to machines at the moment of reliance. Today it is a PDF, a paragraph in a master service agreement, or an opaque number in a vendor portal. None of these is acceptable for a system in which an agent must, at the moment of accepting a transaction, verify that its counterparty is covered, that coverage extends to the specific risk in question, that the policy is in force, and that a defined settlement path exists.

P1 to P5. One question.

Every AIPS-1 Policy Certificate must satisfy five core properties. They roll up to a single architectural question: is this policy credibly enforceable? A Certificate that satisfies all five is one whose existence, scope, trigger conditions, settlement path and current status can be verified by any third party without trusting the Issuer's word for any of them.

Three tiers. By standing-behind, not underwriting quality.

AIPS-1 defines three tiers of Policy Certificate, distinguished by the category of Issuer and the presence of reinsurance backing. The tier reflects the resilience of the standing-behind, not the quality of the underwriting — a well-written Tier I policy may provide better cover for a given risk than a poorly-written Tier III policy. The tier MUST be declared on the Certificate and is verifiable at runtime.

One Certificate. Five properties. Milliseconds.

A Relying Party — counterparty agent, exchange, regulated institution, regulator — verifies the Policy Certificate at the moment of relying on the underlying coverage. The verification flow is designed to be executable by an autonomous agent without human intervention, composing with AIS-1 identity resolution, AES-1 enclave checks (where applicable), and AAS-1 record emission.

// AIPS-1 Policy Certificate verification

const issuer  = await ais1.resolve(cert.issuerDid);
const auth    = await vc.resolve(cert.issuerAuthorisation);

assert(auth.verifySignature());
assert(recognisedJurisdictions.includes(cert.issuerJurisdiction));
assert(aips1.validateCoverageSchema(cert.coverageScope));
assert(aips1.evaluableTriggers(cert.triggers));
assert(cert.settlementPath.endpointReachable());
assert(cert.status === 'Active');
assert(new Date() < new Date(cert.periodEnd));

// Tier III — verify reinsurance backing
if (cert.tier === 3) {
  assert(aips1.reinsuranceAttested(cert));
}

// AES-1 conditional — verify enclave currency
if (cert.aesConditional) {
  assert(await aes1.enclaveCurrent(cert.aesConditional));
}

// Scope alignment with the contemplated risk
assert(aips1.scopeCovers(cert.coverageScope, contemplatedRisk));

// All checks passed — coverage is verifiable and active
return accept(cert);

Reference verification implementations — JavaScript client library, Solidity on-chain verifier, MCP server-side module — ship in v0.2 along with the did:aips1 DID method specification.

Part of an open agent infrastructure stack.

AIPS-1 is designed to compose with the rest of the open agent infrastructure stack. An agent's full runtime attested footprint includes one AIS-1 Bond (identity), optionally one AES-1 Enclave Certificate (execution environment), one or more AIPS-1 Policy Certificates (insurance), and a stream of AAS-1 records (activity).

From working paper to deployed standard.

Part of an open agent infrastructure stack.

AIPS-1 is one of an open family of agent infrastructure standards published by Kadikoy Limited under CC0. Each addresses a distinct subject — identity, execution, audit, remittance, and now insurance.

Specification and supporting materials.

AIPS-1 v0.1 is a draft for public comment.

Feedback is invited from insurance supervisors (IAIS members and national regulators); commercial insurers and reinsurers; captive insurance markets and managers (Bermuda BMA, Cayman CIMA, Vermont DFR, Guernsey GFSC, Singapore MAS); insurance brokers and risk advisers; AI agent developers; legal, regulatory and compliance professionals; standards organisations (ISO TC 68, IAIS, IFRS Foundation); and the wider AI-agent developer community. The comment period closes 30 November 2026. A revised v0.2 will incorporate substantive feedback.

Share your feedback on AIPS-1.

All submissions are reviewed by the AIPS-1 team and will inform the v0.2 revision. The comment period closes 30 November 2026.