AI Agents Could Soon Prove Their Identity and Pay — Without Big Tech
Your AI assistant may soon act for you — this decides who's really in charge.
AI agents are software that can do things for you — book flights, buy supplies, hire other AI helpers, talk to companies on your behalf. Right now, there's no clean way for those agents to prove "I really work for Jane" across different companies. Today that proof lives in separate, centralized systems: one company manages identity, another handles permissions, another handles payment. When you revoke your assistant's access or change your password, everything breaks in the middle — which is exactly when you most need it to hold.
A new paper, NostrAgent, proposes doing all five jobs — identity, permission limits, trust between strangers, discovery, and payment — on one open network called Nostr (a decentralized social network, like email but without a single company running it). Each agent signs its own identity declaration with a digital signature (a tamper-proof stamp only that agent can create), and every time it passes a task to another agent, the permission list must visibly get smaller. A shared trust graph, built from agents vouching for each other, makes it costly to flood the network with fake identities. Payments ride on Bitcoin's Lightning network (fast, cheap crypto payments) so a server can bill an agent directly.
Why does this matter to you? Because you're about to hand real spending power to software. If your AI assistant can buy things, its identity and its spending limits need to be as auditable as a credit card statement — and verifiable even offline, without calling a company's servers. The prototype passed 17 of 19 failure tests, verified identities in under a millisecond, and settled payments in 157 milliseconds. Those numbers matter: proof that being your own "trust authority" doesn't have to be slow.
The honest catch: this is a research paper, not something you can install today. It depends on cryptocurrency rails many people distrust, and the authors disclose one architectural limitation they haven't solved. Treat it as a preview of the plumbing, not a purchase.
- AI agents that act and spend on your behalf currently have no shared way to prove who they work for — NostrAgent proposes a fix using open, signed records instead of corporate gatekeepers.
- In prototype tests, identity checks took under 1 millisecond and payments settled in about 157 milliseconds, with 17 of 19 failure scenarios handled correctly.
- The catch: it's an academic prototype built on crypto rails (Nostr and Bitcoin Lightning), with one admitted design flaw — so no consumer product yet.
Why It Matters
As AI starts spending your money and acting for you, this decides whether you or a big company controls the proof.