METR audit: Anthropic's AI progress may have doubled in speed by July

The rate at which Anthropic's models improve has roughly doubled since October 2025, and about 80 percent of that acceleration traces back to AI being used in Anthropic's own research labs. That's the central finding of nonprofit auditor METR, which spent eleven weeks inside the company and concluded that Anthropic may have crossed its own AI acceleration warning threshold as early as July 15, 2026.

The stakes are straightforward. Whether AI speeds up the development of better AI is treated as a key early warning signal in the safety debate. The concern is a feedback loop: models improve their successors faster and faster, which METR warns could lead to a "sudden emergence of extreme and risky capabilities." Anthropic's own Responsible Scaling Policy names exactly this kind of acceleration as a risk indicator — if the pace of AI progress doubles because of automated AI research, that should trigger a warning level.

What did METR actually measure?

METR studies, among other things, whether AI developers could lose control of their systems. From August 12 to October 29, 2026, a team of 2.5 full-time staffers got what they describe as unprecedented access: benchmark results from unreleased models, training metrics, employee surveys, internal analyses, and roughly 15 interviews with Anthropic researchers. METR took no money from Anthropic but worked on Anthropic devices for security reasons.

The measurement rests on a composite metric called the AECI, which bundles many benchmarks together. The numbers are stark:

• Since October 2025, Anthropic's model test scores have climbed about 1.9 times faster than they did between March 2024 and October 2025.
• Before October 2025, models gained about 15 AECI points per year; since then, the rate has hit 29.
• Anthropic achieved in one year what previously took two — roughly comparable to the jump from Claude 2.1 to Sonnet 4.5.
• The analysis included three unreleased models that finished training between July and October 2026.

METR is confident the acceleration isn't driven by faster growth in training compute. Two years of progress packed into one, powered mostly by AI itself.

How much is AI really doing inside Anthropic?

The auditors estimate about 80 percent of the current speedup comes from AI being used in Anthropic's own research. The employee interviews make the mechanism concrete. One researcher told METR: "Now only bottlenecked on ideas & compute, no longer bottlenecked on code."

Others reported that experiments which used to take one to two days to set up are now "~free." In informal Slack polls from October 2026, employees said on average that the latest internal model made them roughly twelve times more productive. METR treats these self-assessments as "very loose evidence," calling them unrepresentative and likely to overstate the actual gains.

AI isn't doing fully autonomous research yet. The acceleration comes mainly from AI acting as an assistant — but METR spotted "signs of life" that independent contributions could play a bigger role soon. The auditors found no bottlenecks that would slow the pace over the next twelve months, which they say means there's a "significant risk" the acceleration will keep growing over the coming year.

The takeaway: AI is currently an accelerant for human researchers, and nothing METR saw suggests the accelerator lifts anytime soon.

How uncertain are the numbers?

METR itself stresses how imprecise the results are. The true acceleration factor could be as low as 1.3 or as high as 3. And the industry-wide picture looks calmer: when models from OpenAI, Google, and xAI are factored in, the number drops to just 1.4. METR suggests that's possibly because Anthropic lagged behind competitors in 2025, and a catch-up effect is skewing the figures.

Views inside Anthropic aren't uniform either. One researcher pegs AI's share of the productivity gains at just 15 to 20 percent and distrusts colleagues' self-reports. Another considers new hires the single most important factor of recent months. The audit's headline number, in other words, is a point of dispute even among the people being audited.

Why does the evidence look thin?

Anthropic had unilateral redaction rights over all non-public information, and it used them heavily. According to METR, the majority of the evidence the auditors rated as most informative was stripped out — including forecasts for future models, breakdowns by training steps, and the bottleneck analysis. Anthropic describes the redactions as "aimed at removing confidential information that could accelerate other AI developers."

The consequence: the public has to rely on METR's judgment "much more than would be ideal," the organization writes, and any talk of transparency around the evidence is off the table. One graph in METR's report features the words "FAKE DATA" in large letters where the underlying numbers should be.

For an audit meant to give outsiders confidence, the redactions cap how much confidence outsiders can actually place in it.

Did Anthropic cross its own red line?

In a companion review, METR examined Anthropic's risk report from August 2026. In that public report, the company stated the acceleration was "less than a factor of 2." The internal, unredacted version implies a value of 1.97 — and Anthropic itself calls it "plausible" that the threshold has been crossed. METR considers it possible the warning level was already reached on July 15 and accuses the public report of understating both the acceleration and the uncertainty around it.

METR currently shares Anthropic's assessment that the acute catastrophic risk is "low." But the organization can't rule out another sharp acceleration over the next six months, which could quickly push the risk rating higher.

The context matters here. Anthropic CEO Dario Amodei had recently committed to bringing external auditors into the company, and in October Anthropic announced a collaboration with METR involving "employee-like access." The current audit is a separate project with a narrower scope, according to METR.

What happened to the compute audit?

METR was originally supposed to also verify how Anthropic splits its compute between capability research and safety research. Anthropic canceled that part of the project midway through, saying it wanted to focus on risk assessment first and revisit the compute audit later.

That cancellation drew a sharp reaction. Daniel Kokotajlo, the researcher behind the "AI 2027" scenario, argued on X that auditing the compute allocation is the foundation for making commitments to slow down verifiable in the first place.

Without that audit, the public can see how fast Anthropic's AI is improving — but not how the company balances speed against safety. If the acceleration METR detected continues to compound over the next year, the pressure to resume that compute review will only grow.