Anthropic's artificial intelligence systems are now being built at roughly twice the speed they were a year ago, according to an independent assessment released Thursday, and much of that acceleration appears to be driven by AI helping to build more AI. The finding comes from METR, a nonprofit evaluator that conducted an approximately 11-week review of Anthropic's internal data, staff, and research process.
METR staff conducted the high-access evaluation at Anthropic from August 12 through October 29, examining whether the company's AI research and development had sped up and, if so, why, according to METR. Anthropic gave the group what it described as unprecedented access to internal data, including training metrics, benchmark scores, AI speedup analyses, forecasts, and staff interviews. The organization estimated that Anthropic's model capabilities on well-specified tasks were improving at around twice the rate seen between March 2024 and October 2025, though it cautioned that the estimate carries substantial uncertainty.
On a narrower benchmark measure, the group put a sharper number on it: scores grew at an estimated 1.9 times the pre-October 2025 pace, with a 95 percent confidence range of 1.5 to 2.8 times. As reported by 10News, Anthropic's benchmark progress over the past year was roughly equal to its progress between October 2023 and October 2025, according to the METR report. Business Insider adds that METR estimated Anthropic made roughly two years of historical progress over the last year alone.
AI Building AI, Not Just Faster Chips
The group said it was confident the speedup was not the result of faster growth in training compute, ruling out one of the most obvious explanations for a sudden leap in capability. Instead, METR judged that a majority of the acceleration was likely attributable to Anthropic's extensive use of AI to do its own R&D, while stressing that this conclusion remains uncertain. METR researchers attributed much of the faster development to AI assistance rather than AI working entirely on its own, according to Business Insider, describing the dynamic as acceleration achieved primarily via AI assistance rather than fully autonomous contributions.
Anthropic has said its Claude chatbot increasingly helps develop newer and more advanced versions of itself, per 10News. Still, the same report notes that AI was not yet capable of making highly significant, fully autonomous contributions to the most important stages of model training, a limit confirmed by the Herald Sun's coverage of the assessment. The 10News report adds that Anthropic researchers described significant signs that fully autonomous AI contributions to training could arrive soon, a detail attributed to the METR findings.
No Bottlenecks Found, and a Warning About What's Next
METR said it identified no bottlenecks, such as compute or data restrictions, that were holding back Anthropic's AI progress, and it warned that the absence of such limits raised the risk of even faster acceleration ahead. Business Insider reports METR flagged a significant risk that further capability progress could compound, while acknowledging it might be missing important bottlenecks of its own. Separately, informal polls of Anthropic employees suggested productivity improvements of roughly 12 times, but the Herald Sun's reporting notes that METR explicitly cautioned those figures were unreliable and should not be read as a verified measurement.
METR also flagged a possibility that could cut the other way: its findings might overstate real-world impact if the acceleration is concentrated in tasks that are automatically or language-model-graded and well-specified, rather than the messier, harder-to-define work that makes up much of research, according to 10News. The group additionally said Anthropic's public report from August may have understated the overall acceleration now underway.
Redactions Limit What Outsiders Can Verify
Anthropic redacted a majority of the evidence METR considered most informative for its risk-related conclusions, according to METR's own account. The company decided not to share the exact drivers of the speedup because it considers them commercially sensitive and worried they could serve as a roadmap for competitors seeking faster development, per 10News.
Anthropic co-founder and chief architect Sam McCandlish responded to the report Thursday, saying its conclusions largely match his own assessment and that rapid AI progress carries important safety implications that are difficult for outside observers to judge, according to 10News. McCandlish acknowledged that the redactions make it harder for others to independently check METR's work, but said the company plans to release its own analysis of the pace of AI progress within Anthropic in the coming weeks. He noted METR's work began well before, and separately from, any embedded third-party evaluators at the company.
A Debate Amodei Himself Started
The assessment lands against the backdrop of warnings Anthropic chief executive Dario Amodei issued in September 2025, when he called for the AI industry to slow its pace so that safety measures could keep up, per 10News. Amodei defined recursive self-improvement as AI's growing ability to build the next generation of itself, and warned that an unchecked version of that cycle could outrun humans' ability to understand or control the resulting systems. At the same time, Amodei has championed external evaluators embedded within leading AI labs and has pointed to METR specifically as an appropriate third-party reviewer, while OpenAI chief executive Sam Altman has likewise voiced support for independent evaluators with employee-like access.
METR says it does not accept compensation for its work and is funded through donations, and it has evaluated model capabilities for Anthropic, OpenAI, and other labs for years, according to 10News. Partner companies have provided the group with access and tokens for evaluations, research, and engineering rather than direct payment. The assessment arrives as Anthropic prepares for its Wall Street debut, with the company planning to go public later this month in what could become the biggest initial public offering ever, per 10News's reporting.
