Two of the field’s most influential researchers, Geoffrey Hinton and Yoshua Bengio, joined senior figures from OpenAI and Anthropic to warn that an AI “intelligence explosion” could arrive rapidly and upend existing safeguards. In a report co-authored by more than 20 experts, the group argues that automating AI research and development could compress years of progress into months, enabling systems capable of recursively improving themselves with minimal human input.
The authors urge governments to act before capabilities outpace oversight, recommending mandatory, transparent R&D progress reporting with embedded independent auditors; mechanisms to slow or pause development, including datacenter-based controls; strict isolation of automated AI R&D systems; and government-led emergency response plans. They caution that powerful models could accelerate biological and cyber threats, erode human control as people are removed from the loop, and shift geopolitical power if early movers gain decisive advantages.
OpenAI and Anthropic have agreed to independent evaluations of their models; Anthropic says AI now writes 80% of its own code, and OpenAI uses autonomous agents in model training. While the report notes uncertainties—supply-chain limits and regulation could slow deployment—it concludes preliminary evidence supports the possibility of a software-driven intelligence surge and calls for urgent preparation at the highest levels of government.
Related articles:
NIST AI Risk Management Framework (AI RMF 1.0)
OECD AI Principles
