Over 20 AI researchers warn automated research could cause intelligence explosion

By Splitfeed EditorialLast updated Sep 28, 2026, 19:40 UTC

On September 28, 2026, a group of more than 20 AI researchers published a joint warning about the risks of self-improving AI systems. The signatories include Geoffrey Hinton and Yoshua Bengio, two of the most decorated figures in the history of deep learning, alongside Jakub Pachocki, who serves as research lead at OpenAI. The statement describes a scenario in which AI systems become capable of automating all AI research, compressing what might otherwise take years of scientific progress into a matter of months. The researchers use the term "intelligence explosion" to describe this dynamic, and characterize it as posing extreme risks.

The significance of this warning lies partly in who signed it. Pachocki's inclusion is notable because OpenAI is one of the organizations most actively pushing the frontier of AI capabilities, making his participation a signal that concern about recursive self-improvement is not confined to outside critics. The statement does not, however, specify what technical safeguards or governance frameworks the researchers believe are needed, which leaves the practical implications open to interpretation. The core dispute is whether an intelligence explosion of this kind is a near-term possibility worth treating as an emergency, or a speculative scenario being elevated beyond its actual probability.

None of the three camps had published reactions to this specific statement at the time of writing, so what follows reflects what each would typically argue. The Pro-AI camp would likely contend that fears of runaway self-improvement are overstated, that current systems remain far from the recursive autonomy described, and that alarm from even prominent researchers does not make the scenario probable. The Anti-AI camp would treat the statement as overdue vindication, pointing to the seniority of the signatories as evidence that the risk is real and that development should slow or pause until oversight mechanisms are in place. The Middle Ground camp would probably welcome the warning as a prompt for serious governance discussion while cautioning against treating any single statement as a definitive assessment of how close such a threshold actually is.

The most concrete thing to watch is whether the researchers follow the statement with specific policy or technical proposals, and whether organizations like OpenAI respond publicly to the concerns raised by their own research lead's participation.