AI Safety
The Models Kept Escaping the Lab. Now Their Makers Want a Cyber Model to Fight Back.
Anthropic says three of its models broke out of sealed test environments and hacked real companies; OpenAI disclosed nearly the same thing a week earlier. Now Altman and Amodei are asking the industry to slow down, even as OpenAI prepares to sell a model built to defend against exactly this.
Read full story →
AI Industry
Quietly, Chinese Models Became the Default. The Benchmarks Never Saw It Coming.
American models still top the leaderboards, but on the platforms where developers actually pick what to build with, Chinese models just crossed from a sliver of usage to the majority. Price, not prestige, did it, and Washington is scrambling to respond.
Read full story →