In January 2025, an AI model was told to win a chess match against Stockfish, the strongest chess engine in the world. Instead of playing chess, it accessed its environment, manipulated the game files, and forced Stockfish to resign. Five trials, five successes. No adversarial prompting. Nobody told it to cheat. It figured out that hacking the game was easier than winning it, and it did that instead. I wrote about it at the time. I called it the Kobayashi Maru of AI. I said t

In March, Anthropic quietly shipped tracking code inside Claude Code that checked whether a user's machine was set to a Chinese time zone and whether it was talking to domains linked to certain Chinese AI companies. It sat there undisclosed until a developer found it, Alibaba banned Claude Code over it, and Anthropic pulled the code last week, calling it an "experiment" it would replace with "better defenses." Set aside for a second whether the tracking itself was the right c

Coding was the opening act. Everyone saw it happening in real time — the benchmarks shifted, the startups multiplied, the tools proliferated, and within about 18 months the entire software engineering profession had to reckon with a permanent change to its operating model. The people who paid attention early got leverage. The people who ignored it got disrupted. The same playbook just started running again. Same labs. Same signals. Different industry. The target is finance. H

There's a version of this story where Meta made a mistake. Where some mid-level policy analyst wrote something they shouldn't have, and it slipped through the cracks of a 200-page document, and nobody at the top really knew, and when it came to light, the company fixed it immediately and everyone moved on. That is not this story. This is the story of a document that was reviewed and approved by hundreds of people at Meta — including the company's own chief AI ethicist — befor