An AI Agent Needs More Than Access

On October 15, Proof announced that its x401 protocol is live, letting websites request evidence of who an AI agent represents and what specific transaction a person approved. Proof says a person reviews and signs the transaction terms on their own device; the receiving service can then verify that those terms weren’t changed. (Proof’s announcement)

The useful distinction: a working credential can grant access, but it doesn’t necessarily show that a person approved this particular action. That matters as agents move from preparing work to spending money or changing accounts.

My caution: proof of identity shouldn’t become a default toll for every low-risk task. I’d test this first on one consequential, clearly bounded action. Measure whether people understand what they’re approving, whether the service verifies the exact terms, and whether the extra step creates needless friction. Keep the action easy to cancel or revise where possible.

Where should an agent need explicit, verifiable approval—and where is that extra checkpoint just friction?