AI Labs Want In-House Auditors — But Maybe They Should Shut the Front Door First
Dario Amodei proposed embedding outside auditors inside frontier AI companies, and OpenAI, Google and SpaceXAI rallied behind it. Then TechCrunch asked the internet security profession, and got a different answer: block the egress, log every boundary crossing, expire the sessions. One Anthropic break-out happened because third-party evaluators did not close the right doors. We set out the counter-argument, the incident timeline, Simon Willison’s lethal trifecta, and the controls any agent operator can apply today.