Research
I work on establishing trust between parties that share no authority. For people, and increasingly for the AI agents acting on their behalf. Everything below is work I have shipped or published, followed by the questions it left open.
Cross-ecosystem trust interoperability
Every mature identity wallet ships a static list of issuers it trusts, compiled at build time and updated by release. I replaced that list with a live, fail-closed call to a trust registry in eight third-party wallets: the EUDI reference Android wallet, MOSIP's Inji, Talao AltMe, Sphereon, Paradym, Procivis One, SWIYU from the Swiss Confederation, and 2060.io's Hologram. MOSIP runs as a full issuer, wallet and verifier triangle; the EUDI reference wallet is validated on testnet against France Identité's Unfold playground. SWIYU is partial, its positive and control flows are accepted on Android but one verifier request path is still blocked. The integration surface everywhere is the same resolver HTTP API, which is what makes the approach portable rather than bespoke.
Trust registries in production
Verana is a registry of trust registries: a public ledger recording which entities may issue or verify which credential schemas inside a given ecosystem, and the validation process that grants those permissions. I build the frontend and the credential tooling, including JSON Canonicalization Scheme canonicalization, full BCP-47 language tagging, and validation rules read from the chain rather than hardcoded. 182 merged pull requests since February 2026. The open standards counterpart is the Trust Registry Query Protocol, approved as a ToIP specification in December 2025, which defines conformance targets but still has no conformance test suite.
Agent identity and delegation
Autonomous agents raise the same question earlier and harder: when an agent calls a service, who is it, and who authorized it to act? Credat is my open-source answer, a TypeScript library giving agents verifiable identities via did:web and did:key, scoped delegation credentials, and a mutual trust handshake, with adapters for HTTP APIs, LangChain and the command line. The Claw4S protocol formalizes the handshake: two previously unknown agents reach mutual cryptographic trust in exactly two round trips, under 2ms of overhead, producing an auditable chain of signed credentials and a structural guarantee of impersonation detection.
Open Questions
- 1.Trust registries and ETSI trusted lists make different governance assumptions. Where exactly do they fail to meet, and what is the smallest bridge that preserves both?
- 2.A live trust call is a network dependency. What is the correct behaviour when the registry is unreachable, and can fail-closed survive contact with real user expectations?
- 3.How should a wallet surface a trust verdict to someone who has never heard of a trust registry, without either hiding the decision or overwhelming them?
- 4.What revocation mechanisms preserve holder privacy while keeping a verifier's answer fresh enough to be worth asking for?