Anthropic Accelerates AI Development Despite CEO Calls for Caution
Independent analysis reveals Anthropic’s model development speed has doubled, driven by recursive self-improvement.
1Development Speed Doubled
An independent analysis by the nonprofit Model Evaluation and Threat Research (METR) indicates that Anthropic is developing AI systems twice as fast as last year. This assessment, based on eleven weeks of internal data access, suggests a significant acceleration in the company's ability to improve model capabilities for well-specified tasks.
2CEO’s Contradictory Stance
The findings emerge despite CEO Dario Amodei’s public calls for the industry to slow down. In a September 12 essay, Amodei urged AI labs to decelerate development to allow safety measures to catch up, highlighting concerns that rapid progress could outrun human control and lead to potentially catastrophic, unintended consequences.
3Recursive Self-Improvement Mechanism
Researchers believe the primary driver of this acceleration is the use of AI itself to build newer, more advanced versions. Anthropic’s Claude chatbot is increasingly assisting in the development process, a phenomenon known as recursive self-improvement, which allows the company to bypass traditional bottlenecks that typically constrain the pace of innovation.
4Unprecedented Research Access
Anthropic granted METR researchers unprecedented access to internal data and staff interviews ahead of the company's planned public debut. While the report contains caveats regarding potential overestimation and the focus on well-specified tasks, it remains a rare, transparent look into the internal R&D processes of a major AI laboratory.
5Independence and Transparency
The report is significant because technical details regarding AI risks have historically been provided primarily by the companies themselves. With the current administration encouraging self-policing, independent evaluations like those conducted by METR are becoming critical for understanding the safety and trajectory of advanced models before they are deployed at scale.
6IPO Context and Timing
This disclosure comes as the San Francisco-based company prepares for a potential initial public offering later this month. Market observers suggest this could be one of the largest IPOs in history, making the internal development speed and safety governance of the company a central focus for prospective investors and regulators.
7Limits of Current Autonomy
While the report notes significant signs of life regarding AI autonomy, METR clarified that AI is not yet capable of making fully autonomous contributions to the most critical parts of model training. However, the lack of current hardware or data bottlenecks suggests that the compounding effects of self-improvement could accelerate progress.
8Future Safety Implications
Amodei has previously championed the idea of embedding external evaluators within leading AI labs to ensure safety. As Anthropic moves toward its public listing, the tension between rapid technological advancement and the need for rigorous, third-party safety oversight remains a defining challenge for the company and the broader AI industry.