A report co-authored by Nobel laureate Geoffrey Hinton and computer scientist Yoshua Bengio warns that automated AI research could trigger an intelligence explosion, rapidly compressing years of technical progress into months or less. The Guardian reported the warning on 28 September 2026.
The proposed trigger
The authors focus on recursive self-improvement: AI systems using their capabilities to improve the systems that follow them. They argue that automated AI research and development is the most plausible route to such an acceleration because improvements could be deployed quickly once developed.
The paper includes more than 20 authors, among them Anthropic co-founder Jack Clark and OpenAI chief scientist Jakub Pachocki. Its central scenario remains a possibility rather than an established forecast. The authors say current productivity gains from AI-assisted research have not yet reached the level needed to start an intelligence explosion, although newer systems may be approaching it.
Why the authors are concerned
The report says systems capable of expert-level AI research could give a single developer access to a workforce comparable to millions of leading human researchers. That could accelerate scientific and technological progress, but it could also reduce the role humans play in directing or constraining AI development.
The authors identify risks including biological and cyber threats that could advance faster than countermeasures, weaker human control over increasingly capable systems, and governments converting a modest lead in areas such as cyberspace into a decisive advantage. They also acknowledge that supply chains, regulation and other practical constraints could slow the effects, and that AI could speed up risk mitigation as well.
Policy measures proposed
The report calls for transparent progress reporting on AI research and development, including independent auditors inside companies. It also proposes limits on the speed at which AI systems can improve, arrangements with datacentres to pause selected research projects, isolation for automated research systems and emergency plans for possible outcomes.
Evidence from current systems
Anthropic says AI produces 80% of its own code, while OpenAI uses autonomous agents in work such as training new models. These are company statements reported by The Guardian, not independent measurements. The paper also says AI could automate research projects that would take humans months by 2028, while stressing that the timing and consequences of any intelligence explosion are uncertain.
AI-generated editorial illustration; not a documentary photograph.
Source: AI godfathers warn of runaway ‘intelligence explosion’
SIGNAL TITANS
Leave a Reply