An independent research group says Anthropic is improving its AI models at roughly twice its earlier pace, and that most of the speed-up comes from Anthropic's own AI helping to build the next generation. The assessment, published on 5 November 2026 by the nonprofit METR (Model Evaluation and Threat Research), is one of the first outside checks on a question the industry has debated for years: is AI now speeding up its own development?
What METR found
METR staff carried out what they call a high-access evaluation inside Anthropic from 12 August to 29 October 2026. Their headline finding is that Anthropic's benchmark scores, measured on AECI, Anthropic's version of the Epoch Capabilities Index, have grown at an estimated 1.9 times the pre-October 2025 rate, with a 95% confidence interval of 1.5 to 2.8 times. METR reads that as roughly two years of historical progress in the last year. It also stresses the uncertainty: the true figure could be closer to 1.3 times or 3 times, and its robustness checks gave a wider range of about 1.1 to 2.9 times.
METR reports that Anthropic's most advanced internal model, called Model 4 in the report, scored above 50% on every AECI benchmark on which it was tested. Anthropic's internal benchmarks are increasingly saturated, which makes them a less reliable measure of progress as models improve.
How much is AI itself?
METR estimates, with substantial uncertainty, that AI accounts for perhaps around 80% of the current acceleration. It says the acceleration is likely driven mostly by AI assisting human researchers, and that AI cannot yet make highly significant, fully autonomous contributions to the most important parts of model training, according to coverage of the report.
The disagreement about Anthropic's safety threshold
Anthropic's Responsible Scaling Policy sets a threshold related to a doubling of overall progress that can be attributed substantially to AI automation. The 1.9 times figure measures benchmark scores, not that broader quantity, so it does not by itself show the threshold was crossed. Anthropic's filing says it has not designated any current model as exceeding the threshold. METR, however, called Anthropic's argument that the threshold had not been crossed overstated, according to TechTimes. Separately, METR reviewed the automated-R&D section of Anthropic's August risk report.
Much of the evidence is hidden
Anthropic redacted much of the evidence behind METR's conclusions on attribution, bottlenecks and the research pipeline, which METR says means the public must rely on its judgment. Anthropic's Sam McCandlish defended the redactions as commercially sensitive, according to FourWeekMBA, and announced that Anthropic will soon publish its own analysis of its AI progress and Responsible Scaling Policy. AP and Reuters both described the access METR received as unusual for an outside review of a frontier lab. Anthropic declined further comment in some coverage.
Why it matters
If AI is speeding up AI research, progress could compound, which raises both hopes for faster scientific and economic gains and concerns about whether safety measures and oversight can keep up. The report is careful rather than alarmist: it describes assistance, not automation, and gives wide error bars. But because the evidence is partly redacted and the benchmarks are saturating, outside observers will want more data before drawing firm conclusions. Things to watch include Anthropic's promised report, whether other labs allow comparable access, and whether the measured acceleration holds up in later data.
Anthropic is the company that makes Claude, the AI behind this assistant. This article reports METR's published findings and press coverage of them, and I cannot independently verify Anthropic's internal information.
Sources: METR, "Independent Assessment of AI R&D Acceleration at Anthropic in October 2026" (5 November 2026); FourWeekMBA; Crypto Briefing; TechTimes; DEV Community; CellCog.
