Kenny's framing of the "handoff ambiguity" problem is one of the clearest public articulations I've seen of something that's been building for a while particularly the point that traditional security models assume a human explicitly granting access, while autonomous agents need something built for dynamic, unattended delegation instead.
Worth flagging one distinction, though, since the article's proposed fix leans heavily on DNS-anchored identity: proving who an agent is isn't the same problem as proving what it's allowed to do.
A verified "digital passport" tells the receiving system which corporate entity stands behind an agent but it doesn't, by itself, constrain what that agent can actually execute once the handoff happens. Identity and authorization are genuinely separate primitives, and conflating them is exactly how you end up with a fully-authenticated agent still doing something nobody actually approved.
That's the specific gap I've been working on since February a patent-pending framework (UK IPO, GB2603013.0) built around a human-approved, cryptographically signed "intent certificate" issued before an agent starts acting: declared scope, explicit denied actions, hard expiry, a kill threshold, deny-by-default outside that scope, and every gate decision logged to an immutable audit chain.
It's less about proving who the agent is and more about making it structurally impossible for it to act outside what was actually authorized which is where I think Kenny's "Accountability Vacuum" point really bites: without that boundary enforced before execution, the audit trail after the fact just tells you how the disaster happened, not how it was prevented.
Genuinely good piece this problem is getting the serious attention it needs.
Happy to compare notes with anyone else working on the authorization side specifically (not just the identity layer) : IntentBound.com if useful.
