A new research paper lays out a way for AI agents to prove their identity, delegate tasks, and pay for services without relying on a central company to vouch for them.
The system, called NostrAgent, runs on Nostr, the decentralized social protocol built on relays and cryptographic keys. It defines three event types: one for identity declarations signed with BIP340 Schnorr signatures, one for delegation chains that can only narrow an agent's permissions as they pass from agent to agent, and one for peer attestations that build a trust graph resistant to fake-identity floods. Payment rides on Lightning's HTTP 402 standard, tying each transaction back to a specific agent identity. The researchers tested a Python prototype with threat modeling, benchmark suites, and 19 defined failure scenarios, reporting sub-millisecond offline verification and Lightning payments settling at a median of 157 milliseconds.
This addresses a real gap. As agents start hiring other agents and paying for API calls on their own, most current designs assume a platform sits in the middle handling identity and billing, the way OAuth and app stores do for humans. That works until a key needs rotating or a permission needs revoking, and there's no single owner of that process. Rooting identity in cryptographic keys instead of a registry is the same bet Nostr itself made for social media, now extended to machine-to-machine commerce.
It's a prototype with a mixed-method evaluation, not a live network, and two of nineteen failure modes still needed analytical bounding or outright disclosure as limitations rather than clean passes. Worth watching, not worth wiring into production yet.