What the investigation found
AI safety research organization METR carried out an 11-week investigation inside Anthropic. The report shows the company's models are improving roughly twice as quickly as a year ago, with no clear sign of that pace slowing.
In short, Anthropic is using AI to build more powerful AI, and the approach appears to be working remarkably well.
Why it matters
The findings are presented as some of the clearest evidence yet of AI automating its own research and development. Researchers worry that this acceleration could outpace safety safeguards.
A separate program
The probe was separate from a program offering safety researchers "employee-like" access to systems following high-profile hacking incidents over the summer. This report is not part of that access program.
