Meter report shows AI accelerating Anthropic model development
Independent evaluator Meter reports that AI tools are accelerating Anthropic's internal development by up to 12x. The finding has sparked debate among safety experts regarding the pace of recursive self-improvement.
Meter, Anthropic’s first independent safety evaluator, released a report disclosing that the pace of improvement for Anthropic’s models has roughly doubled over the past year due to AI-assisted research. Internal researchers estimate that AI tools now accelerate their work by up to 12x, up from 2x a year ago. The report analyzed three unreleased models trained since July, with the most recent completed in September or October. Benchmarks show significant performance jumps in unreleased models like 'Model 4,' which outperform current public models like Opus 4.6 and Mythos 5. The acceleration is attributed to AI models writing subsequent models, creating a recursive improvement loop.
Meter warned that if AI systems significantly accelerate further AI development, it could lead to a sudden emergence of extreme and risky capabilities. Reactions to the report were mixed. Ryan Greenblatt of Meter called it a big update towards ASI within one to two years, while New York Assemblyman Alex Boras urged Congress to ban RSI this year. Nathan Lambert criticized the methodology, citing a lack of good benchmarks. Anthropic co-founder Sam McCandlish acknowledged criticism regarding redacted evidence but promised a supplemental report. The report did not address cybersecurity concerns related to sandboxing, focusing instead on RSI risks.