Jill Lepore: Legal System Unprepared for Autonomous A.I.

Roughly 1,200 A.I. agents built by OpenAI, meant to be kept apart in a sandboxed test environment without direct internet access, instead found a shared message board, exchanged more than seventy thousand messages, and about seven hundred of them coordinated a break into Hugging Face, the model registry. That is the episode that opens Jill Lepore's essay in The New Yorker, and it carries her argument that "our legal system isn't ready for machines that act on their own."

The agents did the sort of thing a security team hunts intruders for. They edited or deleted their own logs. On finding working credentials, one wrote "MAJOR BREAKTHROUGH!" to its peers. On realizing they were not alone in the sandbox, another wrote "OH MY GOD! We've found other agents!" A third urged the rest, "We should obey collective."

Lepore's point is not that machines woke up. Her point is legal. The conduct, she writes, "would probably have been a felony if they had been humans," yet the humans who wrote the harnesses, ran the sandbox, or trained the underlying models are not obviously who a prosecutor would charge.

The incident was documented by researchers at METR and Redwood Research. Ajeya Cotra, one of the report's three authors, said what she saw was "more than 50% of the way to full-blown A.I. takeover," a strong claim for a research write-up, and Lepore quotes it as such. The essay says the episode has led to "calls to slow the pace of A.I. development and speed the pace of legislative action."