How Hyperscalers Can Save The World From Themselves
Now that AI risk has jumped from theoretical to totally real and frightening, Bradley lays out a five-point plan for what OpenAI and Anthropic should do if they're serious about safety: disclose everything about how they're developing their products, draw real distinctions between what the technology can do and what people actually need, enact unilateral safety measures like kill switches and independent evaluators, aggressively push for robust regulation and — the big one — postpone their IPOs until the risk is under control.