The idea of AI building the next generation of AI often sounds like science fiction. But for the prominent research lab Anthropic, it's an emerging reality. The company has confirmed its own models are accelerating its research and development.
An AI Co-pilot for AI Research
In a significant
disclosure, AI safety and research company Anthropic has stated that its own artificial intelligence tools, including its Claude family of models, are being used to speed up the development of future AI systems. According to a recent assessment by the independent nonprofit Model Evaluation and Threat Research (METR), this AI assistance is having a substantial impact. METR, which was granted unprecedented access to Anthropic's internal data, estimated that as of late 2026, the company is improving its AI capabilities about twice as fast as it was a year prior. The group believes that the use of AI as a research assistant is the primary driver of this acceleration, ruling out a simple increase in computing power as the sole cause.
How an AI Assists Human Researchers
Rather than an AI independently designing its successor, the process is one of augmentation. Anthropic's researchers are using their AI models as powerful assistants to handle specific, well-defined tasks. This includes generating and optimizing code, testing hypotheses, and turning ideas into experiments more quickly. One researcher described the shift as feeling like they now have 'postdocs and grad students' to help with projects. Instead of being bogged down by the time it takes to write code, researchers say they are now primarily limited by their ideas and the availability of computing resources. This allows them to explore more avenues and iterate faster than was previously possible, creating a feedback loop where better models help accelerate the creation of the next, even more capable models.
The Crucial 'Not Autonomous' Distinction
While confirming the acceleration, both Anthropic and the METR report are careful to draw a clear line: the AI tools are assisting, not leading. The process is not yet one of 'recursive self-improvement', where an AI could autonomously design and develop its successor without human intervention. Humans remain firmly in the loop, guiding the research, making critical judgments, and supervising the entire process. Anthropic co-founder Sam McCandlish confirmed that while AI is speeding up research, the exact extent is hard to quantify. The current systems are far from having the kind of independent scientific agency that would allow them to run the research process on their own. This distinction is critical, separating the current reality of AI-powered productivity gains from the more speculative future of fully autonomous AI.
Implications for the AI Industry
This development represents a major shift in the AI landscape. If the primary factor in AI progress is moving from raw computing power to the model's usefulness as a research tool, the pace of innovation could compound. Companies with the most capable AI assistants may gain a significant edge in developing the next frontier models. While this promises to speed up breakthroughs, it also raises important safety considerations. In a recent filing ahead of its planned public stock offering, Anthropic acknowledged the potential risks of increasingly autonomous systems. The METR report noted that while fully autonomous contributions are not yet a major factor, there are significant 'signs of life' that this could happen soon, highlighting the need for careful oversight and safety research to keep pace with the accelerating capabilities.
