Current and former AI researchers are warning that leading labs are racing toward self-improving AI systems before the world is ready, in interviews published by Reuters on Tuesday.

Researchers interviewed by Palisade Research argued the industry is not prepared for models that can recursively improve themselves with little human involvement. DeepMind scientist Neel Nanda put the chance of AI-driven human extinction at 10 per cent, calling that “ridiculously high.”

The warnings come as Anthropic’s Dario Amodei published an essay urging the industry to “pace the frontier”, with OpenAI’s Sam Altman agreeing. OpenAI also said it held back an even more powerful model release.

But critics say the pledges do not go far enough. Former OpenAI researcher Geoffrey Irving said: “If you’re doing a very dangerous thing, you should just slow down.”

The debate lands as both companies race for customers and capital. Several prominent researchers separately published a paper urging policymakers to scrutinise recursive self-improvement work.