An independent assessment of Anthropic's research operations has found that the company's AI models are advancing at roughly twice the pace of a year ago, with most of that acceleration now attributable to AI itself. The report, published Thursday by the AI safety research organization METR after an 11-week investigation inside Anthropic, offers some of the clearest evidence yet that AI is speeding up its own development. The findings arrive as Anthropic prepares to go public, even as CEO Dario Amodei has publicly urged labs to slow AI development.
According to METR, benchmark scores suggest Anthropic packed roughly two years of historical progress into the past year. The researchers said they are confident the speedup was not merely the result of scaling up computing power, which historically has driven AI progress. Instead, they concluded that the majority of the acceleration as of October is due to AI, a contribution they said is growing over time. Slack polls and employee interviews backed that up: staff using Anthropic's most capable internal model reported being about 12 times as productive as they would have been without AI assistance, roughly double what they reported a few months earlier while using Mythos 5.
For now, most of the acceleration comes from AI assisting human researchers rather than working independently, though METR warned there are signs that fully autonomous contributions will play a larger role soon. The researchers said they found no clear signs of diminishing returns or bottlenecks that could prevent even faster acceleration. Much of the gain is tied to unreleased internal models: METR said Anthropic finished training three unreleased models over the past six months, codenamed Model 2, Model 3, and Model 4, completing training in July, August-September, and September-October, respectively. METR also identified an unreleased model dating to 2022, called Unreleased 2022.
Anthropic gave METR access to confidential research, model performance data, and employee interviews, and by METR's account shared more than initially scoped for the project. But the access was not unrestricted, and Anthropic redacted a majority of the evidence METR found most informative from the roughly 100-page full report, according to METR. METR said its information pipeline was relatively narrow and that some data requests were denied, and it noted the public must now rely on its judgment more than would be ideal. Cofounder Sam McCandlish acknowledged on X that the company redacted a lot and accepted that this makes it harder for others to check METR's work, saying Anthropic judged that sharing exactly what is driving the speedup is commercially sensitive and would hand rivals a roadmap to go faster.
The report has drawn fierce criticism from AI safety researchers and commentators, who point to a contradiction at the heart of Anthropic's approach: the company has been among the most outspoken advocates for safety and responsible development, yet the findings suggest it is helping accelerate the technology toward a threshold many consider dangerous. Some observers said the results suggest that even aggressive forecasts about AI progress may be too conservative, and that recursive self-improvement, in which AI systems autonomously develop more powerful successors with little or no human involvement, may be closer than previously believed. The report covers only Anthropic, though OpenAI said in September that it is making strong progress toward building an AI researcher that can further progress on deep learning and alignment by March 2028.
More AI news from TechManNews.
