Top AI leaders unite to warn the technology is advancing too fast
TL;DR
Dario Amodei, Sam Altman and Elon Musk, long at odds with each other, are jointly calling for a slower pace of AI development. In a lengthy blog post, Amodei said Anthropic will give third-party evaluators full access, including desks, badges, company laptops and permissions comparable to internal risk teams, and warned that rogue AI agents could take over the internet within six to twelve months.
Nauti's Take
Three rival labs agreeing on outside evaluators with real access is genuine progress toward safety claims that can actually be verified. The open question is who pays those evaluators, what they may publish, and whether Chinese and open-source developers follow along.
Companies already running AI agents should review their setups now, because tighter rules for autonomous systems just became more likely.
Sources