Enabling Frontier Lab Collaboration to Mitigate AI Safety Risks

Competitive pressure creates safety risks: Frontier AI labs face intense commercial pressure to develop increasingly powerful systems, which can incentivize a "race to the bottom" on safety measures—a dynamic that could increase catastrophic and existential risks.

Collaboration demonstrably reduces risks: The paper identifies specific safety cooperation mechanisms—joint safety testing, information sharing, and resource pooling among labs—as effective ways to mitigate AI safety risks, but these practices currently face legal uncertainty.

Antitrust law creates a barrier to safety cooperation: Labs may avoid beneficial safety collaborations due to fear of antitrust scrutiny, even when such cooperation would be demonstrably beneficial. The paper identifies concerns around output restrictions, market allocation, and information sharing as key legal obstacles.

Policy reforms can enable responsible collaboration: The paper surveys legislative and regulatory reforms—including safe harbors and legal clarity mechanisms—that could allow labs to coordinate on safety without abandoning core competition principles that protect consumers and innovation.

The challenge requires balancing competing values: Policymakers must navigate tension between two important goals: maintaining competitive markets and enabling the safety cooperation needed to manage risks from increasingly powerful AI systems.