Anthropic AI development speed doubles as models begin aiding their own growth

Accelerated growth rates

Anthropic is currently developing its artificial intelligence systems at roughly double the speed it maintained a year ago. This finding comes from an analysis by Model Evaluation and Threat Research (METR), an independent nonprofit based in Berkeley, California. The group was granted unprecedented access to Anthropic's internal data and staff interviews to conduct this review.

The report highlights that Anthropic has made as much progress on benchmarks in the last year as it did between October 2023 and October 2025. This rapid advancement is occurring even as the company prepares for a major initial public offering on Wall Street. The findings are particularly notable because they come from an outside group rather than the companies themselves.

METR noted that while the data was partially redacted by Anthropic, they could confidently identify the trend. The researchers focused on whether the growth rate of internal capabilities had accelerated and how much of that speed was attributable to AI. They concluded that the company's rate of improvement has indeed doubled during the assessment period.

The role of recursive AI

One of the most significant findings in the METR report is that AI is increasingly playing a role in its own development. Anthropic has stated that its Claude chatbot is helping the company build newer and more advanced versions of the software. This process is often referred to as recursive self-improvement, where a system uses its own capabilities to create its successor.

The researchers were able to rule out increased computing power, or compute, as the primary reason for this jump in performance. Traditionally, the amount of hardware power used in training is the main factor that determines how well a model performs. Because compute was not the driver, the researchers believe the acceleration stems from the efficiency of using AI to improve AI.

However, the report includes some caveats regarding the scope of these improvements. The acceleration might be limited to well-specified tasks that can be graded automatically by a language model. It may not yet apply to more complex or messy tasks that are harder for researchers to define or measure accurately.

Safety and oversight concerns

The rapid pace of development creates a complex situation for Anthropic's leadership regarding safety. CEO Dario Amodei has recently urged the industry to slow down so that safety measures can keep pace with technological gains. He expressed concern that unchecked advances could outrun the ability of humans to understand or control these systems.

Amodei specifically advocated for the use of external evaluators to monitor the progress of leading AI labs. He highlighted METR as an appropriate third-party organization to provide this level of oversight. This move toward independent auditing is supported by other industry leaders, including OpenAI CEO Sam Altman.

The situation is complicated by the fact that the federal government has encouraged AI companies to police themselves on safety matters. By providing internal data to a nonprofit like METR, Anthropic is participating in a form of self-regulation. The goal is to identify potential bottlenecks or risks before the technology becomes too advanced to manage.

Looking toward the future

While AI is not yet capable of making fully autonomous contributions to the most important parts of model training, researchers see signs of life in that direction. The METR report suggests that the lack of bottlenecks in data or computing power could allow this acceleration to compound. This creates a scenario where the speed of improvement could increase exponentially as the models become more capable.

Anthropic's upcoming public debut will likely bring more scrutiny to these development speeds and the risks associated with them. Investors and regulators will need to weigh the benefits of rapid progress against the potential for catastrophic consequences. The company must balance its drive for innovation with the safety protocols Amodei has championed.

Moving forward, the industry will likely see more reports from independent groups like METR. As AI companies continue to use their own models to build the next generation of technology, the focus will shift toward how to maintain human control. The transition from human-led development to AI-assisted development marks a significant shift in the lifecycle of artificial intelligence.