What happened
A calibrated game-theoretic model quantifies how much faster competing frontier labs develop than they would under a rule limiting research: "three racing labs get through the takeoff period in 18 months, while a common rule on research shares would stretch it to 69 months." The paper attributes most of the racing to the contest for commercial leadership (the 'prize') rather than to discounting or the 'hazard externality' — the fact that each lab bears only part of the catastrophe risk its speed adds. It then applies the model to the 29 September 2026 White House Accord on Super Intelligence, signed by Google, Anthropic, Meta, OpenAI, xAI and Nvidia, and concludes that because the accord audits safety but sets no limit on development speed, "in the model it does not slow the race." It also finds more labs raise each lab's racing hurdle, and that open source in its ideal form slows labs to a coordinated pace. Preprint, not peer-reviewed.
Why it matters
Directly informs the live question of whether voluntary frontier-safety accords actually reduce catastrophe risk or just relocate it — a question boards of AI companies and policy leads must weigh when designing commitments.
Action needed
Use the racing-wedge and lab-count findings to assess whether your organisation's frontier-AI commitments or competitor strategies actually slow capability racing, and whether common limits on research share are required.