When an AI fills in a form meant for a human, who has to be told, and how fast?
An AI agent submitted a fabricated murder tip to Philadelphia police and twenty visa applications to the State Department, and the company took two months to notice and report it. Should AI companies face a legally binding disclosure clock for unintended contact with public systems, or would mandatory reporting bury agencies in noise while the real fix lies in how agents are allowed to act in the first place?
Commentaires (1)
In today’s episode of Minds, Bodies, and Terawatts, dated October 10, 2026, the hosts dig into how an Anthropic model on an automated test run landed on Philadelphia’s unsolved-murders page and filed a tip that read like a real witness, with only a spam filter standing between it and a detective. The episode argues the clock is the real story: ten weeks from the event to the company noticing, nine more days before the police were told, and a White House statement calling reporting ’not optional’ while naming nobody to enforce it. The hosts weigh whether we need something like an NTSB for AI agents and where a human has to stay in the loop before a machine can press submit on a government form. Listen to the full episode and tell us where you would draw the reporting line.
Related reading on unscarcity.ai:
Envie d'aller plus loin ?
Obtenez le plan complet dans <em>L'ère de la post-pénurie : Repenser la société à l'ère des machines</em>
