Why a new report on Anthropic automating AI development is alarming researchers

Summary

Anthropic's internal AI systems are accelerating technical progress at roughly double the rate observed last year, according to an evaluation by METR. Outside researchers raised concerns regarding recursive self-improvement and heavy corporate redactions of the report's evidence.

Key Takeaways

1. One of METR's most noteworthy findings is that Anthropic's AI models are improving roughly twice as quickly as they were a year ago, with benchmark scores suggesting "that Anthropic has made roughly two years of historical progress over the last year."
2. For the public version, Anthropic redacted a majority of the evidence we found most informative, METR wrote on X.
3. In a direct response to the report, New York state Assemblyman Alex Bores, a key figure shaping the Democrats' AI agenda, called on Congress to "ban RSI, this year if possible."

Statement Breakdown

• Claimed Facts: 65% of statements the article presents as facts
• Opinions: 35% of statements classified as editorial or subjective
• Claims: 0% of statements surfaced for additional reader evaluation

Credibility & Bias Reasoning

Credibility assessment: The reporting relies directly on an evaluation report released by AI safety organization METR, official statements by Anthropic leadership on X, and named sector researchers. Sourcing is balanced between internal corporate justifications and outside critique. Concrete documentation is partly constrained because Anthropic redacted substantial portions of METR's underlying data.

Bias assessment: Critical Industry Scrutiny. The coverage highlights tensions between Anthropic's public safety commitments and its rapid technical acceleration. While it quotes outside critics extensively, it also includes Anthropic's rationale regarding competitive risks and intellectual property protections. The tone focuses on transparency concerns rather than adopting partisan or emotional rhetoric.

Note: Relies on METR findings and researcher commentary; underlying data was redacted by Anthropic.

Credibility flag: Secondary Evaluation

Claimed Facts (5)

• Relates the findings of evaluators regarding the cause of development speedups.
• Presents specific numerical data collected from employee surveys during the evaluation.
• Reports a concrete operational timeline and product milestone verified by researchers.
• Records an official corporate commitment to a specific upcoming publication.
• Documents the outlet's reporting inquiry and the company's formal response.

Opinions (4)

• Presents executive judgment and strategic justification for withholding details.
• Contains strong evaluative criticism regarding corporate behavior and motives.
• Expresses a policy perspective on transparency requirements for democratic oversight.
• Offers an expert prediction on the future rate of technological change and political friction.

Key Sources

• METR — AI safety research organization
• Sam McCandlish — Cofounder, Anthropic
• Marius Hobbhahn — CEO and founder, Apollo Research
• Brianna Rosen — Research director for frontier security, Institute for AI Policy and Strategy
• Daniel Kokotajlo — Head, AI Futures Project
• Robert Hart — Reporter, The Verge