AI firm speeds up development twice as fast
Independent researchers found Anthropic is improving its models at double the rate of a year ago, despite CEO calls to slow down.
By Martin Reynolds ·
An independent research group reports that Anthropic is developing artificial intelligence systems at roughly double the speed of a year ago.
The nonprofit Model Evaluation and Threat Research, based in Berkeley, California, released this finding on Thursday. The analysis contradicts recent public statements from Anthropic CEO Dario Amodei, who urged the industry to slow its pace of improvement. Amodei published an essay on Sept. 12 calling for a reduction in the rate of capability gains to allow safety measures to catch up. He described this push as a response to mounting concerns about runaway AI systems.
METR gained access to Anthropic’s internal data to conduct its assessment. Researchers reviewed information from mid-August through late October, a period spanning about 11 weeks. They also interviewed company staff and examined both public and private risk reports. The group stated that the public report released by Anthropic in August understated the level of acceleration occurring within the company. METR noted that redactions in the private data limited its ability to share detailed estimates of the speed increase.
The study suggests that AI itself is driving this faster development. Researchers ruled out increased computing power as the primary cause of the acceleration. Instead, they attributed the speedup to the use of AI tools to improve other AI models. Anthropic has stated that its Claude chatbot increasingly assists in developing newer versions of itself. METR found no major bottlenecks, such as data restrictions or compute limits, holding back progress. However, the report noted that AI is not yet making fully autonomous contributions to the most critical parts of model training. Anthropic researchers described early signs that this capability could arrive soon.
Sam McCandlish, an Anthropic co-founder and responsible scaling officer, responded to the report on the social platform X. He agreed with many of METR’s conclusions, noting that it is difficult to quantify exactly how much AI is speeding up research. McCandlish explained that the company withheld specific details about the drivers of the speedup because the information is commercially sensitive. He added that sharing the exact process might provide competitors with a roadmap to accelerate their own work.
Anthropic is preparing to go public later this month. The company plans to hold an initial public offering that could become the largest stock offering in history. The timing of the report coincides with the company’s move toward the stock market. METR, which is funded by donations, has evaluated models for Anthropic, OpenAI, and other labs for several years. The group does not accept compensation for its evaluations, though partner companies provide access and computing tokens.
The independent assessment highlights a shift in how the industry monitors itself. President Donald Trump’s administration has encouraged AI companies to police their own safety standards. Amodei and OpenAI CEO Sam Altman have both supported the idea of embedding independent evaluators within leading AI labs. METR was specifically named by Amodei as an appropriate third-party group for this role in his September essay.
