OpenAI Concedes No Path to Safe Full RSI; xAI Targets 2027

Claude now leads 26% of Anthropic's model research and development, completing most tasks 'end-to-end from a high-level prompt' under human supervision, Fortune reported on September 19.

That is the concrete number in a story otherwise built on named executives saying, on the record, that they are approaching a threshold they cannot yet make safe. OpenAI acknowledged it does not yet know how to 'safely get all the way to aligned, full RSI,' and cannot assume 'progress in alignment and safety will keep pace.' The company is building an automated 'research intern' capable of tasks that take 'a skilled researcher a few days,' with a target of an autonomous AI researcher by March 2028.

Elon Musk's xAI is more direct. Musk said 'humans are gradually getting less and less in the loop' on Grok model improvement, and the company targets full automation by the end of 2027.

The critic Fortune builds the story around is Anthony Aguirre, president and CEO of the Future of Life Institute and a physics professor at UC Santa Cruz. 'I think this is probably the worst idea in the history of humanity to do this,' he said. He also defined the term: autonomous recursive self-improvement, in his words, means 'AI that can improve itself designing the next version of the system, then the next version, and so on.' The compounding is the point. 'The really important thing here is that as AI is doing more of it, it gets faster, because AI operates just much, much more quickly than the humans do.'

The counterweight from inside academia is not that RSI is far off, but that it is already here in a partial form. John Thickstun, an assistant professor of computer science at Cornell, told Fortune: 'We have already, for years, been using these models in supportive roles for creating the next version of these models.'

Anthropic said it would slow or pause development if global competitors did so 'in a verifiable manner.' The piece sits inside a busy week of safety coverage on our tracker, and describes no verification mechanism actually in place.