An independent assessment by AI safety organization METR reveals that Anthropic's AI models are improving nearly twice as fast as historically recorded, driven largely by internal AI-assisted research and development rather than increased compute scaling. The findings raise concerns about recursive self-improvement and AI alignment risks, coinciding with Anthropic CEO Dario Amodei's call to pace frontier development as the company prepares for an initial public offering.
METR study finds accelerated Anthropic AI development
- ▪METR based its 11-week assessment of Anthropic, conducted between August and October 2026, on confidential research, employee interviews, and performance data from unreleased models
- ▪A November 5, 2026 METR study found Anthropic's AI models improve on well-specified tasks about twice as fast (1.5 to 2.5 times their March 2024 to October 2025 rate) with high uncertainty, not driven by increased training compute
METR evaluation timeline and Anthropic response
- ▪METR began communicating with Anthropic regarding its risk assessment in mid-August 2026 before submitting its final public report on November 3, 2026
- ▪Anthropic did not immediately respond to requests for comment regarding METR's assessment of Anthropic's accelerating AI capabilities published on November 5, 2026
Internal AI use drives Anthropic productivity gains
- ▪METR reported that Anthropic's capability acceleration is likely driven by its extensive internal use of AI tools to perform research and development tasks
- ▪Anthropic researchers reported sharp productivity gains from using AI to write code and conduct experiments, with informal internal employee polls suggesting productivity improvements of about 12 times
Risks and limits of AI self-improvement
- ▪AI researchers warned that recursive self-improvement could make advanced AI systems difficult to monitor, control, or align with human goals, potentially posing an existential threat
- ▪METR's findings offer evidence that AI systems are accelerating their own development, potentially paving the way for recursive self-improvement without human intervention
- ▪METR clarified that AI systems are not yet capable of making highly significant, fully autonomous contributions to the most critical stages of model training
Story comments
Loading comments…