AnswersPublished14 sources
METR Finds Anthropic’s AI Research Progress Is Moving About Twice as Fast
METR judged that Anthropic was improving models on well specified tasks at about twice its earlier rate. In a separate review, METR agreed that direct catastrophic risks from Anthropic’s July 15 models were low, while saying the company may already have crossed its early warning threshold for research acceleration.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- METR judged that Anthropic was improving models on well specified tasks at about twice its earlier rate.
- In a separate review, METR agreed that direct catastrophic risks from Anthropic’s July 15 models were low, while saying the company may already have crossed its early warning threshold for research acceleration.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did nonprofit METR’s independent, high-access evaluation of Anthropic from August 12 to October 29 reveal about AI accelerating its own. Article summary: METR’s evaluation points to AI materially accelerating the development of successor AI systems—not yet to a demonstrated, fully autonomous self-improvement loop. Its separate risk-report review distinguished low immediat. Topic tags: general, news, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
A high-access review by the nonprofit Model Evaluation and Threat Research (METR) found that Anthropic’s progress on well-specified AI tasks was likely moving about twice as fast as it had from March 2024 to October 2025. METR’s assessment points to AI helping accelerate AI research, but it is a high-uncertainty judgment—not evidence that systems are autonomously building their own successors. 8
What METR’s evaluation found
METR evaluated Anthropic from August 12 to October 29, 2026, using access to internal research and information about the company’s development process. Its top-line conclusion was that Anthropic was likely improving model capabilities on well-specified tasks at roughly twice the earlier rate. 8
The report also described a benchmark-specific estimate: since October 2025, Anthropic’s scores on its capability index had grown at about 1.9 times the pre-October 2025 rate, with a 95% confidence interval of 1.5 to 2.8. That statistical interval belongs to the benchmark estimate; it should not be treated as the full uncertainty around METR’s broader judgment. METR said the true acceleration could plausibly be between 1.3 and 3 times the historical pace. 8
METR tentatively estimated that AI accounted for a majority of the current acceleration—perhaps around 80%. The organization presented that figure as a guess, not a precise measurement of cause and effect. 41
What the findings do not establish
The results describe faster progress on well-specified tasks. They do not, on their own, show that AI systems are autonomously conducting all or most of the work needed to train more capable successors. The finding is better understood as evidence of AI-assisted acceleration than proof of a self-sustaining feedback loop. 8
The public evidence also has limits. METR said Anthropic permitted it to publish its highest-level conclusions but redacted much of the evidence it considered most informative. That makes the assessment important, while limiting how fully outsiders can inspect the basis for it. 8
41
Why METR’s separate risk review matters
A separate METR review examined Anthropic’s August 2026 risk report and its assessment of models deployed internally by July 15. METR agreed that direct catastrophic risks from those models were low. But it argued that Anthropic’s case for saying its early-warning “2x Acceleration Threshold” had not been crossed was overstated, and said the threshold may plausibly have been crossed.
Those conclusions are not contradictory: low direct catastrophic risk from the models assessed is different from concern that AI-assisted research could speed up future development. METR’s warning concerned the pace of research acceleration, not a finding that catastrophe was imminent.
The connection to Amodei’s call to pace AI development
In September, Anthropic CEO Dario Amodei called for slowing the pace of AI capability improvements and proposed embedded third-party evaluators, including organizations such as METR, to assess companies’ safety practices and commitments. 13
16 METR’s high-access assessment illustrates the kind of outside review that proposal describes, while also showing how redactions can constrain public scrutiny.
8
16
The clearest conclusion is therefore an early warning, not a declaration of runaway self-improvement: METR sees likely acceleration in AI development, believes AI assistance may be a major factor, and emphasizes substantial uncertainty about the evidence and its implications. 8
41
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "METR Finds Anthropic’s AI Research Progress Is Moving About Twice as Fast"?
METR judged that Anthropic was improving models on well specified tasks at about twice its earlier rate.
What are the key points to validate first?
METR judged that Anthropic was improving models on well specified tasks at about twice its earlier rate. In a separate review, METR agreed that direct catastrophic risks from Anthropic’s July 15 models were low, while saying the company may already have crossed its early warning threshold for research acceleration.