The race to build more powerful artificial intelligence has entered a new phase. AI firm Anthropic is now using its models to help design the next generation of AI, a process that is dramatically speeding up its research and development.
What is AI-Assisted Research?
For years, building
AI models was a painstaking, human-led process. Now, AI is becoming the researcher's most powerful assistant. At Anthropic, this means its AI model, Claude, is involved in nearly every stage of development. It helps run research experiments, prepares and cleans training data, evaluates new models, and even writes the vast majority of the company's code. An independent assessment by the research organisation METR found that this AI assistance is the primary reason for a major acceleration in Anthropic's progress. This method, sometimes called recursive self-improvement, involves delegating a growing share of the development cycle to the AI systems themselves, marking a significant shift from traditional AI development.
A 'Dramatic Acceleration' in Progress
The results of this new approach are stark. According to the METR report, Anthropic is improving its AI models' capabilities roughly twice as fast as it was a year ago. To put that in perspective, the company made about two years of historical progress in the last year alone. This acceleration is not due to simply using more computing power; instead, it is attributed mainly to the growing role of AI in the R&D process itself. In August 2026, an internal analysis found that AI was leading more than a quarter of the work on developing the next Claude model, up from just 1% a year prior. This has prompted Anthropic to warn that AI-assisted research could advance model capabilities faster than it can develop and validate the necessary safeguards.
Why Human Supervision is Crucial
Despite the speed gains, Anthropic stresses that humans must remain in control. The company and its CEO, Dario Amodei, have been vocal about the potential risks of AI that improves itself without oversight. Amodei has warned that if not properly controlled, AI could move beyond human understanding and capabilities. For now, even when an AI 'leads' a task, it is still under human supervision. Anthropic says it does not anticipate allowing its AI to 'fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding'. The company is even considering policy changes to its Responsible Scaling Policy to ensure that people, not algorithms, retain ultimate control over the direction of AI development.
Balancing Speed with Safety
This entire strategy highlights the core tension at the heart of Anthropic, a company founded by former OpenAI researchers with a focus on AI safety. The company is in a competitive race with rivals like Google and OpenAI, and using AI to accelerate its own research provides a key advantage. However, this acceleration brings new safety concerns. Anthropic has acknowledged that this rapid, AI-driven progress could introduce significant safety risks, potentially outpacing the work being done to ensure these powerful systems are aligned with human values. This is why the company has a unique corporate structure designed to protect its safety mission and has publicly committed to precautionary measures, such as delaying the release of some powerful models.
