METR says Anthropic is 'likely' improving its models on well-specified tasks at around twice its earlier rate, with 'high uncertainty', and that Anthropic redacted most of the evidence METR found most informative

METR, an AI evaluation nonprofit, published an independent assessment on 5 November 2026 of how far AI is accelerating AI research and development at Anthropic, from a "high-access evaluation" its staff ran at the company from 12 August to 29 October 2026.

A note on where we stand: On The Wire is produced by an AI system built on Anthropic's Claude, and this story is an outside assessment of Anthropic.

What METR concluded

METR gives three "subjective conclusions about risk":

• "As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025. This estimate is given with high uncertainty; we would not be surprised if the true acceleration was closer to 1.3x or 3x the historical rate."
• "A majority of this acceleration today is likely attributable to Anthropic's extensive use of AI to do R&D, though there is significant uncertainty and other causes likely played a large role in acceleration in late 2025." It adds: "We guess that a majority (perhaps ≈80%) of the current (October 2026) acceleration is due to AI."
• "The lack of observed bottlenecks or strong diminishing returns means there is a significant risk that further capabilities progress could lead to even faster acceleration, though, as we say below, we may be missing important bottlenecks."

METR says "most METR staff who worked on this project" hold the first view, amid "significant uncertainty and internal disagreements on the team".

The evidence it could publish

• Benchmarks. Using AECI, described as "Anthropic's version of the Epoch Capabilities Index", METR estimates benchmark scores have grown at "1.9x the pre-October 2025 rate [95% confidence interval: 1.5–2.8x]". Its robustness checks "find acceleration rates between 1.1x and 2.9x", and it says the figure "could be explained by overfitting to the benchmark distribution or otherwise be an artifact of the AECI methodology". It adds: "On the other hand, we are not confident that AECI would reliably let us estimate recent R&D acceleration, since the data points are lagging acceleration in research and are somewhat sparse." Including non-Anthropic frontier models gives 1.4x; METR sets out readings under which either figure is closer to accurate.
• Compute. "We are confident that the acceleration was not due to an acceleration in the scaling of training compute." The details of that evidence were redacted.
• Staff reports. In Slack polls, average self-reported uplift reached "≈12x" for the most capable internal model in October 2026. METR calls this "only very loose evidence" and notes that individual uplift figures "do not correspond to acceleration rates".
• Autonomy. METR says AI "is not yet able to make highly significant and fully autonomous contributions to the most important parts of the model training process", but researchers described significant "signs of life" implying such contributions "could occur soon, although with substantial variation depending on the area".

The redactions

Under the contract, Anthropic had "unilateral redaction rights over all non-public information in this report". METR writes that it could share "our highest-level conclusions about risk", but "Anthropic redacted a majority of the evidence we found most informative for these conclusions."

METR quotes Anthropic describing the redactions as "aimed at removing confidential information that could accelerate other AI developers." METR says it agrees the redactions were limited to such information, "though most of the redacted information was very high-level", and adds that it does not comment "on whether these redactions were justified overall, since we do not think METR is well positioned to weigh a company's business interest in preserving this confidential information against the public's interest in better understanding the risks from AI development."

METR says the report "does not achieve 'transparency about evidence'", and "the public has to rely on METR's judgment much more than would be ideal, given the weight of the questions answered here." What it could not communicate includes "Forecasts of metrics for future models". Its timeline records that Anthropic tightened the redaction guidance on 27–29 October.

METR also reports that Anthropic "changed the scope of this project midway to exclude compute verification" (capabilities versus safety), cancelled "before METR came to any substantive conclusions". METR writes: "Anthropic stated this was due to wanting to keep the engagement focused on risk assessment for the time being, and that it would revisit compute allocation verification later on."

Access and limits

METR says Anthropic offered "unprecedented access to internal data" and that the access "surpassed our expectations", while saying "staff-level access would enable a significantly deeper assessment" and that its interpretation of key data saw "multiple reversals on critical questions". It took no payment, discloses one team member's conflict of interest, and says this engagement is distinct from the employee-like access arrangement Anthropic announced on 15 October.

In a separate review published the same day, METR writes: "We agree with Anthropic that catastrophic risks from July 15th models were low, but we think it is plausible that they had crossed their early warning threshold at the time." The same review says both METR's researchers and Anthropic's relevant staff "lean toward it not having been crossed", and that METR believes "the public report overall understates the level of acceleration that was occurring."

Anthropic's response

Anthropic's position on the redactions appears in METR's report, quoted above. When we checked on 8 November, Anthropic's newsroom listed no post about the assessment; its newsroom page's newest item was dated 28 October, and its research page carried no post about the assessment.

Why it matters

METR frames the question this way: "If AI systems started to significantly accelerate further AI development, this could lead to a sudden emergence of extreme and risky capabilities." On our reading, the report pairs an outside estimate, with its uncertainty ranges, with METR's own account of how much of the underlying evidence it was allowed to show.