Anthropic’s AI Pace May Have Doubled, METR Finds

An independent assessment published Thursday (Nov. 5) concludes that Anthropic, one of the world's leading AI developers, is probably improving its models about twice as fast as before, and that much of the speedup likely comes from using AI in its own research. Over the same weekend, news reports raised questions about how the United States and Nvidia control sales of advanced chips to companies linked to China. The debate over data centers in the U.S. after the midterm elections rounds out the day. None of the sources address Brazil. Together they show who sets the pace and the rules for AI technology worldwide.

Anthropic: report points to doubled pace of progress, with much left out

METR, an organization that evaluates the risks of AI systems, worked inside Anthropic from Aug. 12 to Oct. 29 with access to internal data and published its findings on its website. As of October 2026, METR estimates, Anthropic is probably improving its models' capabilities on well-defined tasks at about twice the rate recorded between March 2024 and October 2025. METR itself calls the estimate uncertain: it says it would not be surprised if the true figure were close to 1.3 or to 3 times.

Using a benchmark-based capability index, Anthropic's version of the Epoch Capabilities Index (AECI), METR put growth since October 2025 at 1.9 times the earlier pace, with a 95% confidence interval of 1.5 to 2.8 times. That amounts, METR says, to roughly two years of historical progress in the past year. The organization says it is confident the speedup does not come from more computing power in training. It says most of it probably comes from using AI in research, with significant uncertainty, mostly as an aid to researchers, though there are signs that fully autonomous contributions may soon carry more weight. It also says it has seen no clear bottlenecks over the next 12 months, adding that its understanding of this changed several times during the work.

The most sensitive issue, according to METR and The Verge, is transparency. Anthropic had the right to keep information out of the public version of the report and, METR says, omitted most of the evidence the organization considered most informative. Anthropic co-founder Sam McCandlish acknowledged on X that this makes the work harder to check. He defended the decision as the right call, arguing that releasing more could benefit competitors, The Verge reports, and said the company plans to publish its own report on the pace of AI progress in the coming weeks. The Verge also cites internal surveys in which employees using the company's most capable internal model said they were about 12 times more productive on average than they would be without AI. That figure is self-reported and hard to interpret, as METR itself admits.

The report describes internal research and, in part, unreleased models. The discussion it prompts is about how fast AI is advancing and how much the public can verify of what happens inside the labs.

Nvidia: chip sales to China and a new requirement for cloud providers

The New York Times reported, as relayed by The Verge, that the U.S. government approved purchases of Nvidia's H200 AI chips by Chinese companies tied to firms on the government's own restricted list. The example cited is New Matrix: the newspaper reports that the office listed as the firm's address is empty, that administrators of the industrial park say they have no record of it, and that its owners are linked to a Chinese asset manager that has been on a U.S. blocklist since 2024. The paper describes the case as part of a pattern, and The Verge reports concern that neither the Commerce Department nor Nvidia is rigorously checking buyers. The information comes from the newspaper's reporting and, in the sources consulted, has not been independently confirmed.

On another front, the Financial Times, cited by TechRadar, reported that Nvidia has begun requiring cloud providers to earn at least 51% of their revenue from customers outside China as a condition for new contracts. TechRadar adds that company representatives communicated the rule verbally over the summer, and it applies only to future deals but affects customers worldwide. In the same report, Nvidia said that long-standing U.S. policy encourages all cloud providers to build on American infrastructure and to seek customers anywhere. TechRadar also says the company has cut the number of authorized buyers in Asia by more than half this year, and that a rack of B300 processors, sold for about $600.000 outside China, reportedly went for roughly $2.8 million there in October because of shortages.

What the reports show is that U.S. rules and Nvidia's own rules influence the availability of this equipment to cloud providers in many countries.

Data centers: U.S. midterm elections raise local pressure

TechCrunch said in its weekly discussion that the outcome of this week's midterm elections could weigh on the industry. The site reports that in dozens of local ballot measures, voters rejected data centers, and elected officials, including Texas governor-elect Gina Hinojosa, promised to restrict them. The piece carries a correction: the measures on the ballot were largely restrictions or moratoriums on data centers, and almost all of them passed. The participants disagree about how much weight the issue really carried: one says it was not on the ballot everywhere, and another says it was not what drove voters to the polls but has become a grassroots issue. One participant says OpenAI released a policy plan for "compute campuses" that waives nondisclosure agreements with local officials; another participant doubts this will change anything.

Building data centers is one of the conditions for offering AI services, and the debate over where to put them and under what rules is open in other countries. The source does not analyze the Brazilian case.

Read more

Today's summary is at AI Updates on RaisLife. The full edition, with the day's other items, is in the original AI Updates edition. See also the technology page.

Sources

• METR: Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
• The Verge: Why a new report on Anthropic automating AI development is alarming researchers
• The Verge: The US let Nvidia sell chips to an empty office in China
• The New York Times: report on Chinese buyers of Nvidia chips
• Financial Times: report on the requirement for revenue from outside China (cited by TechRadar)
• TechRadar: Nvidia and the new rule for cloud providers
• TechCrunch: After the midterms, things are looking rocky for Elon Musk and data centers

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. Data: technology at RaisLife.