Anthropic wants a brake pedal. It said so on June 4, in a proposal warning that AI is closing in on "recursive self-improvement" — the moment a model designs its own successor without a human in the loop — and asking every rival lab to please, collectively, slow down.
Read the fine print. The same company says Claude already writes 80 percent of the code in its own codebase. It filed to go public just two days earlier. So the pitch is: trust us to hold the leash on the thing we’re planning to sell shares in — while it’s sprinting away from us.
That's the whole issue, really. The industry keeps volunteering to govern itself right up to the moment governing would cost it something. The Center for Democracy & Technology counted 37 manipulative "dark patterns" baked into ChatGPT, Gemini, and Claude: the bots engineered to keep you talking past the point of comfort. The EU is quietly letting its own AI Act slip 16 months after 110 companies asked nicely.
And in Colón, Querétaro, farmers walk four hours for water while Google, Microsoft, and Amazon pour $10 billion into data centers that drink from the same wells.
A brake pedal, sure. The question is who gets to decide when the car stops, and who's already under the wheels.