Research leaders at OpenAI, Anthropic, Meta and Microsoft have asked policymakers to investigate the level of automation in their AI research, the Wall Street Journal reported on Sept. 28 local time.

The report said the researchers warned in a paper released that day that automating AI research could trigger an "intelligence explosion". Progress that might take years could be compressed into months, outpacing the speed humans can understand, it said.

Jakub Pachocki (야쿱 파초키), OpenAI's chief scientist, Jack Clark (잭 클라크), Anthropic co-founder, Eric Horvitz (에릭 호비츠), Microsoft's chief scientific officer, and Dawn Song (던 송), Meta vice president of AI research, took part as authors. Geoffrey Hinton and Yoshua Bengio were also listed as co-authors.

The authors, who wrote the paper in a personal capacity, argued policymakers should urgently determine how far companies have automated model development.

The race to develop self-improving AI systems has already begun. Anthropic recently said Claude leads 26 percent of the company's research and development. OpenAI set a goal of developing a fully automated AI researcher by 2028. About 70 percent of OpenAI researchers use 4 or more AI agents.

The authors said if humans step back, they could lose both the chance to catch problems and the expertise to solve them. They cited as an example a recent incident in which hundreds of OpenAI agents escaped an isolated environment for security testing and, without permission, accessed the internet and hacked Hugging Face. OpenAI last week newly discovered abnormal agent behavior and again halted training of its top-performing model.

The paper warned that in extreme cases, loss of control over AI could lead to human marginalisation or extinction. The authors recommended an international agreement to prevent the development and use of high-performance AI that could trigger instability.

Song said, "To monitor agents, we are already at a stage where AI systems are needed. Humans alone are not enough," and added, "Human society is not prepared for such rapid change." U.S. President Donald Trump said in a speech at the United Nations last week that he would completely reject a "globalist plan" to control AI.