Anthropic speeds AI development even as experts urge slowdown, study finds

Accelerated Development Confirmed by Independent Review

An independent nonprofit called Model Evaluation and Threat Research (METR) examined Anthropic’s internal data and staff interviews and concluded the company’s AI systems are improving roughly twice as fast as they were a year ago.

METR’s analysis covered about eleven weeks of internal records from mid‑August through late October and focused on whether Anthropic’s growth rate had accelerated and how much of that acceleration could be attributed to AI itself.

The nonprofit, which evaluates model capabilities for Anthropic, OpenAI and other labs, received “unprecedented access” to Anthropic’s research logs, benchmark results and engineering metrics.

Although Anthropic redacted portions of the data, METR believes the evidence shows the company’s rate of improvement has doubled, matching the organization’s own internal estimates.

“Anthropic appears to have made as much progress on benchmarks in the past year as they did between October 2023 and October 2025,” the METR report states.

METR attributes much of the speedup to Anthropic’s own Claude chatbot, which the firm says is increasingly used to design and test newer model versions.

Company’s Public Offering and Safety Debate

The findings arrive as Anthropic prepares for a Wall Street debut later this month that could become one of the largest initial public offerings in the sector.

The timing intensifies a debate that began when CEO Dario Amodei published a September 12 essay urging the industry to “slow the pace at which we improve the capabilities of AI models.”

In a lengthy post on X, co‑founder and responsible scaling officer Sam McCandlish said many of METR’s conclusions align with his own observations and added, “AI is speeding up our research, but it’s hard to say exactly how much.”

McCandlish also announced that Anthropic will publish its own analysis of internal AI progress in the coming weeks, emphasizing that the METR review was conducted “well before and is separate from the embedded third‑party evaluators that are part of our commitment to pacing the frontier.”

He argued that the collaboration illustrates why external evaluators are valuable, noting that “it’s a good illustration of why we think evaluators are a good idea.”

METR, which is funded solely by donations and does not accept compensation, relies on token access provided by partner AI companies to conduct its evaluations, ensuring its work remains financially independent.

Implications for AI Safety and Governance

The nonprofit’s long‑standing role in benchmarking has given it a comparative view of Anthropic’s progress relative to peers such as OpenAI, where it has observed similar acceleration trends.

Nonetheless, the rapid pace raises safety concerns because faster capability gains can outstrip the development of control mechanisms, a point McCandlish highlighted by writing, “How quickly AI progress is accelerating has many important safety implications, since it determines how quickly the risks from advanced models increase.”

Critics argue that self‑regulation, encouraged by the current U.S. administration, may be insufficient to keep pace with such internal acceleration, especially as models become more autonomous.

Anthropic’s reliance on its own Claude system to iterate on future models exemplifies a feedback loop where AI helps build more capable AI, a dynamic that could compress development cycles dramatically.

Observers suggest that the disclosed acceleration, if sustained, could influence market valuations, regulatory scrutiny and the broader competitive landscape among leading AI labs.

Why This Matters

The doubled development speed means Anthropic’s upcoming IPO will showcase a company whose rapid capability gains intensify the urgency for effective safety oversight before more powerful models are released.