Anthropic invited independent researchers to assess its AI progress, then redacted their report
Anthropic’s AI capabilities are accelerating despite its own calls to slow the progress.
Image by Samuel Boivin via Shutterstock
Follow us
- Anthropic gave METR unusual access to assess how AI accelerates its own AI research.
- METR found Anthropic’s model capabilities may be improving at about twice the previous rate.
- Researchers said around 80% of the acceleration likely comes from Anthropic using AI in research and development.
- Anthropic redacted much of the report, saying full details could help rivals speed up their AI work.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Anthropic invited third-party researchers to assess how AI is driving its AI research. The company then admitted to heavily redacting the findings so they wouldn’t inform rivals trying to accelerate their models.
AI lab Anthropic gave independent investigators from the Model Evaluation and Threat Research (METR), a non-profit research institute, an “unprecedented” level of access to evaluate the degree to which AI is accelerating AI research and development (R&D).
The METR researchers were able to access internal data, including a variety of pre-training metrics, benchmark scores for internal models, and AI speedup analyses.
They published the report on November 5th, 2026, which includes the following findings:
- 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.
- A majority (around 80%) of this acceleration is likely attributable to Anthropic’s extensive use of AI to do R&D.
- The lack of observed bottlenecks or strong diminishing returns poses a significant risk that further progress in capabilities could lead to even faster acceleration.
- AI is not yet capable of making fully autonomous contributions, but shows significant “signs of life.”
The researchers noted being “confident” that the acceleration wasn’t due to increased scaling of the training compute.
The findings come weeks after Anthropic CEO Dario Amodei and other leaders of major AI labs called for slowing down AI progress.
The calls followed a series of warnings by current and former employees that AI is being developed irresponsibly and could cause the destruction of humanity in the next decade.
In September, Amodei emphasized the need for “pacing the frontier” and suggested a three-step plan for imposing better guardrails, which included granting an “employee-like access” to a team of “third-party evaluators.”
Anthropic doesn’t want others to “go faster”
The METR report may offer a glimpse into how the proposal to involve independent evaluators could work in practice.
Sam McCandlish, a co-founder and chief architect at Anthropic, admitted in an X post that the company “redacted a lot” in the report and acknowledged that this makes it “harder for others to check METR’s work.”
But we determined that sharing exactly what is driving the speedup isn’t the right thing to do, both because it’s commercially sensitive and because it risks handing out a roadmap for others to go faster,Sam McCandlish, a co-founder and chief architect at Anthropic
McCandlish argued that disclosing the exact causes of AI acceleration would further increase the risks METR is worried about, adding that the company “made the right call.”
Meanwhile, the METR team said Anthropic redacted “a majority of the evidence” they found most informative for the conclusions about potential risks.
The disclosure that the findings were redacted drew criticism from industry insiders.
Daniel Kokotajlo, the executive director of the AI Futures Project, criticized Anthropic’s decision to redact important information to prevent others from accelerating AI, because the company “does not appear to be slowing down much at all.
He argued that Anthropic was throwing itself “behind efforts to pace the frontier with nearly the gusto that you would if it was your top priority.”
Kokotaljo wrote, “Not to mention ‘commercial sensitivity,’ you yourself have said human extinction is a plausible outcome. Why are you even talking about commercial factors?”
