Anthropic's AI Is Building Better AI, Report Warns
AI that improves itself could change everything—but is it safe?
A new report from AI safety group METR found that Anthropic's AI models are helping build better AI, and it's working. Over the past year, the company's models improved about twice as fast as before. Employees using the most advanced internal AI reported being 12 times more productive—double the boost from just a few months earlier. Most of this acceleration comes from AI assisting human researchers, but METR warns that fully autonomous AI contributions may soon play a larger role.
Anthropic has already trained three new AI models—codenamed Model 2, 3, and 4—that haven't been released. These internal models are driving much of the productivity gain. The report also notes that Anthropic was allowed to redact parts of the final report, and it did so extensively. While the company shared more than expected, the secrecy raises questions about how much we really know about the pace of AI development.
This matters because AI that can improve itself could lead to rapid breakthroughs—or rapid risks. Anthropic's CEO has publicly called for slowing down AI development, yet his own company is speeding up. As Anthropic prepares to go public, the findings highlight a tension between innovation and safety. For everyday people, this means AI could become more powerful faster than expected, affecting jobs, privacy, and daily life.
- Anthropic's AI is making its own development twice as fast as last year.
- Employees say AI makes them 12 times more productive, but Anthropic kept key details secret.
- Three new AI models are trained but unreleased, hinting at even faster progress ahead.
Why It Matters
Self-improving AI could bring faster breakthroughs but also bigger risks, affecting jobs and safety.