What happened
A report from the research organization METR, following an 11-week investigation into Anthropic, concludes that the company’s AI models are improving roughly twice as quickly as they were a year ago. The findings indicate that Anthropic has achieved approximately two years of historical progress within a single year. METR researchers attributed the majority of this acceleration to AI-driven productivity gains rather than simply scaling resources. The report also revealed that Anthropic has completed training on three previously unreleased models—codenamed Model 2, Model 3, and Model 4—between July and October 2026, alongside an older model from 2022.
The METR report, based on an 11-week internal probe, found that Anthropic's models are improving at double the speed of the prior year. Researchers noted that this acceleration is primarily driven by AI assistance in the development process, with employees reporting a 12-fold increase in productivity when using the company's most capable internal models.
Anthropic has maintained a rapid development cycle, completing three new, unreleased models (Model 2, Model 3, and Model 4) in the latter half of 2026. These models remain internal and are not available to the public.
The assessment process was subject to significant restrictions. While Anthropic provided access to confidential research and employee interviews, it exercised its contractual right to redact a majority of the evidence METR deemed most informative. METR noted that these redactions make it difficult for the public to independently verify the findings or assess the associated risks.
Why it matters
The report provides concrete evidence that AI is increasingly capable of accelerating its own development, a phenomenon known as recursive self-improvement. This trend raises significant concerns among safety researchers regarding the speed at which frontier AI systems are advancing and whether current safety protocols can keep pace. The findings have sparked criticism regarding Anthropic's transparency, as the company exercised its right to redact significant portions of the METR report, citing commercial sensitivity and the risk of providing a roadmap to competitors. This has fueled accusations of hypocrisy, given Anthropic's public advocacy for responsible AI development and calls for industry-wide caution.
The findings suggest that the timeline for recursive self-improvement—where AI systems autonomously develop more powerful successors—may be shorter than previously anticipated. This acceleration challenges existing safety frameworks, which may not be designed to handle the rapid, iterative improvements now being observed.
The tension between Anthropic's public stance on and its internal development speed has drawn sharp criticism. Critics, including former researchers, argue that the company is racing ahead while simultaneously calling for industry-wide caution, creating a perceived contradiction that undermines trust in self-regulation.
The redaction of the METR report has highlighted a broader issue regarding the lack of independent, transparent oversight for frontier AI labs. Experts like Brianna Rosen of the Institute for AI Policy and Strategy argue that without the ability for independent evaluators to share their full findings, democratic oversight and public accountability remain compromised.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
The primary focus is on the upcoming report Anthropic has promised to release regarding the speed of AI progress and new safety measures. Additionally, the industry will be monitoring potential legislative responses, such as calls from policymakers like New York Assemblyman Alex Bores for federal bans on recursive self-improvement (RSI) technologies. Observers are also watching for how other frontier labs, such as OpenAI, progress toward their own goals of building AI researchers capable of advancing and alignment, as well as whether independent safety evaluators will secure deeper, less restricted access to internal lab systems.
Anthropic is expected to publish its own report on the pace of AI progress and new safety initiatives in the coming weeks, which will be scrutinized for how it addresses the concerns raised by the METR investigation.
Legislative pressure is mounting, with some officials calling for explicit bans on recursive self-improvement technologies. The political climate, particularly with new legislative sessions approaching, may lead to more stringent regulatory proposals for AI labs.
The role of independent safety evaluators is evolving. Organizations like Apollo Research are pushing for deeper, embedded access within AI labs to ensure that safety assessments are based on comprehensive evidence rather than limited, redacted disclosures.