AI is taking a larger role in developing the next generation of AI systems, but the leading labs’ accounts do not show machines improving themselves end to end without human supervision. That distinction matters: progress is real, while the runaway feedback loop often imagined in debates about AI remains a future possibility, not an established capability.
Watch Desk analysis
What happened
ABC News reports that Anthropic says Claude now leads 26% of its model research and development, completing much of a given task from a high-level prompt while people remain involved. OpenAI describes an automated “research intern” for defined tasks under human direction and says it aims to build an automated AI researcher by March 2028. xAI chief executive Elon Musk has said people are gradually becoming less involved in improving Grok, while clarifying that the process is not yet fully automated.
Researchers use “recursive self-improvement” in different ways. It can mean AI assisting in the development of better AI, or the much stronger scenario in which systems autonomously design successor systems in repeated cycles. Those are not interchangeable milestones, however convenient it would be if they were.
Why it matters
AI tools helping researchers write code or complete research tasks could speed development. Fully autonomous systems repeatedly improving their successors would be a more consequential shift, and the evidence cited by ABC News does not establish that this has been achieved.
The key practical question is whether human oversight and safety work keep pace as AI takes on more of the research. OpenAI says it does not yet know how to reach fully aligned, full recursive self-improvement safely. Anthropic has said it would consider slowing or pausing development if competitors did so verifiably too.
Our read
The 26% figure is a notable measure of AI’s role inside one lab, not proof that Claude is independently building its successors. Keep the distinction between assistance and autonomy firmly in view; the debate gets less useful when a research tool and a self-directed improvement loop are treated as the same thing.
What to watch
- Whether Anthropic publishes clearer definitions and comparable measures of AI’s contribution to research.
- Whether OpenAI’s “research intern” moves from directed tasks towards broader autonomy.
- What evidence labs provide that human oversight and safety measures are keeping pace with capability.
Discussion spark: What evidence should convince the public that AI is moving from assisting researchers to improving itself autonomously?
Sources and evidence
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.