“A new METR report, released following an 11-week investigation, reveals that Anthropic is using its own AI models to accelerate AI development, resulting in roughly twice the improvement rate compared to a year ago. The findings have alarmed AI safety researchers who warn that self-accelerating AI development could outpace safety measures. The trend shows no signs of slowing, raising urgent questions about oversight and control.”
Key Takeaways
- METR published a report after an 11-week investigation into Anthropic's AI-assisted development practices.
- Anthropic's models are now improving approximately twice as fast as they were one year ago.
- The acceleration shows no clear signs of plateauing, heightening concerns among AI safety researchers.
Anthropic's models are improving at double last year's pace, and researchers are alarmed.
Part of the 7 November 2026 AI news digest
trending_upWhy It Matters
When AI systems actively contribute to building more capable successors, the development cycle can compress dramatically, leaving safety evaluations and regulatory frameworks struggling to keep up. For AI practitioners, this raises immediate questions about whether existing red-teaming and alignment techniques scale fast enough to match the accelerating capability curve. For the broader industry, Anthropic's results may pressure competitors like OpenAI and Google DeepMind to adopt similar self-improving pipelines, potentially normalising the practice before its risks are well understood. Policymakers should watch whether self-accelerating development triggers new disclosure requirements or intervention from bodies like the EU AI Office or the US AI Safety Institute.
FAQ
What exactly is METR, and why does its report matter?
METR is an independent AI safety research organisation that evaluates the capabilities and risks of frontier AI systems. Its reports carry significant weight because METR operates independently of the labs it studies, making its findings a rare outside view into internal development dynamics.
Is Anthropic the first AI company to use AI to build better AI?
No, using AI to assist in model training and research is an increasingly common practice across frontier labs, including OpenAI and Google DeepMind. What makes the METR report notable is the documented scale and speed of the improvement Anthropic is achieving through this approach.
What are the specific safety concerns researchers have raised?
Researchers worry that if AI systems can accelerate their own development, human oversight may not scale fast enough to identify dangerous capability jumps before deployment. The core fear is a feedback loop where each more capable model builds an even more capable successor faster than safety teams can evaluate it.
