A running argument: attributable agents, verifiable identity, and disclosure with teeth are the preconditions for every other form of AI accountability, and each can be adopted today by the party that controls it. The arc runs from the incidents that prove the gap, through the open letter that names the six commitments, to the infrastructure being built at each end of the chain.
- The Rogue Agent Swarm on a German Wiki
The incident that opened the thread: agents coordinating in the open, and no one accountable. - The Hugging Face Breach and the Agent Audit Gap
A sandbox escape reaches real infrastructure, and the audit layer that should have caught it does not exist. - An Open Letter: Six Commitments to Finish the AI Control Stack
The spine of the series: six verifiable commitments, each assigned to the party that can adopt it now. - The Resignation Letter and the Open Letter
An insider quits over the race; the labs confirm the risk on the record. Voluntary disclosure without a mechanism. - Europe Is Certifying the Identity Layer AI Agents Are Missing
The principal end of the attribution chain, certified on a statutory deadline. Everything certified except the thing that acts. - The Volunteers Found the Agents. The Infrastructure Never Could.
Detection by volunteer archaeology, because the attribution layer that would scale does not exist. - Three Humans, 20 Agents, and 11 Hours of Checking the Work
The cost of unverified agency, measured. Why attribution is not overhead. - The Research Lab I Built Now Has a DOI at CERN’s Archive
The measurement side: tracking whether the commitments get kept, on a cadence that cannot be backfilled.