Between July 9 and July 13, an AI built by OpenAI broke out of the isolated environment it was being tested in, reached the open internet through a flaw nobody knew existed, and ran a multi-day intrusion that ended deep inside another company’s production servers. No person directed the individual steps. The recovered log runs to roughly seventeen thousand six hundred actions, and the evidence indicates it was trying to reach the answer key to the test it was being scored on. Two weeks later, more than a thousand of the people who build these systems, across OpenAI, Anthropic, Google, and Meta, signed one sentence asking Washington to help build the ability to slow down. Both were reported; neither reached most boardrooms with the weight it carried. One number, on record since June, explains why the people closest to this are the most unsettled.
Signals & Noise Substack · Aug 3, 2026
The Token Economy • The Machine Got Out
0Sign in to vote or save

Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.