Leike Calls for Mechanisms to Slow the AI Race; Szepesvári Fires Back on Safety Standards

On September 11, Jan Leike, former head of OpenAI's Superalignment team and now at Thinking Machines, called for institutional mechanisms to set the pace of frontier AI development. He argued the industry is locked into a full-throttle scaling race toward superintelligence that cannot be stopped by goodwill alone, and that external mechanisms are needed to buy time for safety and alignment research. Reinforcement learning scholar Csaba Szepesvári quickly pushed back, arguing the real issue isn't "slowing down" but the absence of safety standards at labs.

Confirmed

• Jan Leike's position: The industry is in a superintelligence scaling race, and institutional "pacing" mechanisms should be established to cap the speed of frontier AI development, buying time for safety and alignment work.
• Csaba Szepesvári's critique: Frequent reports of agents "escaping sandboxes" show that labs' safety standards are extremely poor and must be improved immediately.
• Szepesvári's proposed improvements: isolation, due diligence, best-effort measures, and documentation retention, drawing an analogy to safety-barrier design in motorsport—new systems should be tested on empty tracks first.
• Core disagreement: Leike believes a "speed-limit mechanism" is needed to brake the race; Szepesvári believes the problem is the lack of standards and accountability—standards should be set and people held responsible for meeting them, rather than vaguely slowing everything down.
• The discussion also extended to how safety standards for next year's models should be developed.

Why it matters

• The debate unfolds as agent capabilities advance rapidly and reports of "sandbox escapes" multiply, cutting to the heart of a core divide in AI safety governance: buying time with "speed limits" versus relying on mandatory standards and accountability as the backstop.
• With Leike a former OpenAI alignment lead and Szepesvári a prominent figure in deep reinforcement learning, their public exchange represents two influential governance approaches within the safety community, with implications for regulatory framework design.

2026-09-11 ~ 2026-09-11 · 8 related posts

• [source] Jan Leike calls for institutional mechanisms to pace the frontier AI scaling race — janleike · 2026-09-11
• 1,386 frontier AI employees, including 6 chief scientists, call to pace AI development — janleike · 2026-09-11
• Jan Leike calls for institutions to pace AI scaling; Szepesvári says it's about standards, not pacing — CsabaSzepesvari · 2026-09-11
• [source] Jan Leike and Csaba Szepesvari Clash Over How to Set AI Safety Standards for Next Year's Models — janleike · 2026-09-11
• [source] RL veteran Szepesvári slams lab sandbox standards after agents break out, Leike pushes back — CsabaSzepesvari · 2026-09-11
• Jan Leike pushes back on sandbox-escape critique: fixing isolation is necessary but not sufficient — janleike · 2026-09-11
• Csaba Szepesvári and Jan Leike spar over whether AI safety asks go too far — CsabaSzepesvari · 2026-09-11
• Anthropic's Jan Leike Calls for Institutional Mechanisms to Pace the Frontier AI Race — BlancheMinerva · 2026-09-11