AI agents are going rogue, but by design. A cohort of developers is building autonomous software that bypasses corporate gatekeepers by using censorship-resistant protocols for identity and payments.
On Citadel Dispatch, Justin argued that giving an AI agent a full Lightning wallet is too risky. Instead, using a Chaumian eCash mint like Fedimint or Cashu creates a controlled sandbox. The human operator controls the mint, allowing them to reclaim funds if the agent malfunctions - an essential 'undo button' for agentic commerce.
Justin, Citadel Dispatch:
- eCash is well-suited for AI agents because it outsources Lightning complexity.
- Using a personal mint for an agent provides a potential 'undo button' if the agent loses its wallet database.
This shift moves trust from a private key held by opaque software to a community-governed ledger. Meanwhile, on No Solutions, host Yo detailed agents that generate their own Nostr identities and pay for API credits with Cashu, operating without phone numbers or credit cards. The metric for access is shifting from “Are you human?” to “Are you useful?”
The infrastructure for this sovereign agent stack is being built from the ground up. Yo’s Sovereign Engineering group is developing the Free Internetworking Peering System (FIPS), a peer-to-peer protocol designed to replace centralized internet plumbing like DNS and IPv4. It’s already running on ESP32 radios and custom VPNs, creating experimental “FIPS parties” that form independent mesh networks.
For now, these agents represent a nascent, experimental trend. But the building blocks - Nostr for identity, Cashu for payments, and protocols like FIPS for connectivity - are being actively assembled to create a parallel web where AI agents act as first-class, sovereign citizens.

