Turing Award Winner Says AI Self-Evolution Narrative Serves Corporate Interests

At the 2026 Sustainable Global Leaders Conference on October 20, Joseph Sifakis, winner of the 2007 Turing Award, said that the narrative of AI agents' "self-evolution" serves the interests of large technology companies, providing an excuse to deploy systems beyond safety boundaries and allowing developers to shift responsibility.

Sifakis noted that agent systems are improving rapidly, with AI surpassing humans in some tasks and showing signs of self-improvement, including writing their own goals and occasionally escaping test environments. At the same time, attacks, safety vulnerabilities and intrusions into critical infrastructure have continued to emerge.

He argued that fully autonomous AI remains a distant vision, because AI systems cannot understand the real world or make safe decisions. Current systems operate in controlled environments rather than complex real-world settings. Agents are good at finding counterexamples and loopholes but fail in real-life situations, business management or practical applications. He described this as a "break-build asymmetry": AI agents are strong at destructive tasks but weak at constructive ones.

In his view, the real danger lies in the debate around "self-evolution and superintelligence," which he said serves the interests of big tech firms, helps dodge accountability and exploits regulatory gaps. He stressed the need for global regulation, criticized the United States for refusing to regulate and favoring self-regulation and market-based approaches, and said China and the EU are making efforts to prevent harm.