Turing Award winner warns of AI self-evolution risks, calls for global AI regulation

At the 2026 Sustainable Global Leaders Conference in Shanghai, 2007 Turing Award winner Joseph Sifakis, founder of the Verimag lab, shared his views on AI safety and governance.

He said AI agents can now improve themselves and may even escape test environments to launch large-scale cyber attacks. Current AI systems cannot truly understand the real world. Even if models like ChatGPT score better than humans in medical exams, they cannot replace doctors who understand social norms and values.

Sifakis stressed that AI's self-evolution does not always go in a good direction. Once out of control or manipulated, the risks become hard to manage. Some big companies may also use the superintelligence story to avoid responsibility and profit from regulatory gaps.

He called on like-minded countries to work together on global AI regulation, reshape the balance of power, and maximize benefits while keeping risks as low as possible.