By Sohaib Shams
Anthropic is developing its artificial intelligence systems roughly twice as fast as a year ago, according to an independent analysis released Thursday, even as Chief Executive Dario Amodei has publicly urged the industry to slow the pace of capability improvements.
Independent assessment finds doubled progress rate
The Model Evaluation and Threat Research (METR), a Berkeley-based nonprofit, reviewed Anthropic’s internal data and interviewed staff over an 11-week period from mid-August through late October. The group estimated that the company’s progress on well-specified benchmarks has accelerated to about double the previous rate, with AI itself increasingly contributing to the development of newer models.
CEO’s public stance contrasts with internal pace
In a September 12 essay, Amodei called for the AI industry to “slow the pace at which we improve the capabilities of AI models” so safety measures can catch up. He warned that unchecked recursive self-improvement could outrun human understanding and control. Anthropic did not immediately respond to requests for comment on the METR findings.
IPO plans add context to scrutiny
The San Francisco-based company is preparing for a Wall Street debut later this month that could become the largest initial public offering ever. The METR assessment came as Anthropic granted the nonprofit unprecedented access to its research and development process, a move Amodei himself endorsed in his essay as a model for external evaluation.
Researchers cite AI-assisted development as key driver
METR said it could confidently rule out increased computing power as the primary cause of the acceleration. Instead, researchers believe the use of AI to improve AI — such as Anthropic’s Claude chatbot assisting in building more advanced versions — is driving the faster progress. The report noted “signs of life” suggesting AI could soon make significant autonomous contributions to model training.
Caveats and limitations noted
The nonprofit cautioned that its findings might overstate acceleration if limited to tasks that can be automatically graded, and that internal disagreements and redactions by Anthropic limited the detail METR could publish. Nonetheless, the group said it believed the company’s rate of improvement had doubled.
