Scientists from leading technology companies and research centers are urging governments to urgently introduce controls on the automation of artificial intelligence development, saying the ability of algorithms to create their own successors could trigger an uncontrolled “intelligence explosion.”
Scientists from leading technology companies and research centers are urging governments to urgently introduce controls on the automation of artificial intelligence development, saying the ability of algorithms to create their own successors could trigger an uncontrolled “intelligence explosion.”
This was reported by Business Insider with reference to a joint article by researchers and materials from a discussion with the participation of Alan Chan, a research fellow at the GovAI think tank.
The letter to policymakers, co-authored by Anthropic co-founder Jack Clarke, OpenAI chief scientist Jakub Pakhotski, Microsoft chief scientific officer Eric Horvitz, and Nobel laureate Jeffrey Hinton, calls for mandatory internal and independent auditors to monitor companies' attempts to automate AI research.
“Imagine, for example, that we encounter Mythos-level events not every year, but every month in new areas. This is the kind of opportunity we call an ‘intellectual explosion,’” Alan Chan explained.
This process, he says, is related to recursive self-improvement (RSI). If models start writing code and conducting research to create the next generation of AI, it will create a feedback loop and accelerate change to an undetermined pace.
Development automation is already a reality: Anthropic reported that as of May, AI systems generated over 80% of the company’s approved code, compared to just a few percent in early 2025. Researchers expect significant or even complete automation of R&D in this area over the next few years.
Chan emphasizes that accelerating technology development can bring benefits—such as accelerating medical discoveries—but it also increases the risks of large-scale incidents that result in loss of control over systems.
The day before, the heads of Anthropic, Google, OpenAI, SpaceXAI, and Nvidia signed an agreement at the White House to cooperate with independent evaluators. Donald Trump called the document "morally binding," but experts note that the agreement lacks specifics, and the regulatory window for implementing effective measures is rapidly closing.
Recall that in June 2026, Anthropic released its flagship models Claude Mythos 5 and Fable 5, designed for deep code analysis and vulnerability detection. Mythos 5 demonstrated the ability to find thousands of zero-day vulnerabilities in critical infrastructure, which is why the model was placed under temporary export restrictions by the US government .
