A research agenda titled Pacing the Frontier has outlined how governments and independent experts might slow the development of increasingly capable artificial intelligence systems. The report, coauthored by University of Toronto researcher Raymond Douglas, describes enforcement as an unresolved technical and political problem rather than a simple agreement among AI laboratories.
Proposed controls include independent evaluations inside frontier laboratories, stronger red-team inspections, monitoring large computing runs, and cryptographically recording activity on advanced chips. Researchers have also discussed international agreements with China and other countries, because restrictions imposed by one government would have limited effect if development continued elsewhere.
Industry leaders disagree on safety controls
Early public support from leaders at Anthropic, OpenAI, Google DeepMind, and SpaceX suggested that parts of the industry might accept coordinated safeguards. That apparent consensus quickly fractured. The Verge reported that Meta chief executive Mark Zuckerberg rejected limits on individual laboratories, while Zuckerberg, Elon Musk, and Nvidia chief executive Jensen Huang helped derail a proposal for an industry-funded independent regulator.
OpenAI continues to support government involvement. Global affairs chief Chris Lehane told The Verge that Congress should establish mandatory national safety standards for frontier AI. President Donald Trump has moved in the opposite direction, dismissing the latest safety warnings and arguing that existing presidential, criminal, and regulatory authority is sufficient.
Read nextOpenAI plans safety report and tighter ChatGPT limits on violence-related contentLabor unions focus on immediate workplace risks
Entertainment labor groups have added a different challenge to the slowdown debate. SAG-AFTRA and the Writers Guild of America East told The Verge that policymakers should not let speculative extinction scenarios distract from current problems involving jobs, consent, compensation, disinformation, and corporate accountability. Major studios contacted for the report did not respond.
SAG-AFTRA chief negotiator Duncan Crabtree-Ireland said the union already secured enforceable rules requiring permission and compensation when studios use digital replicas of performers. He acknowledged broader AI risks but questioned whether some extinction claims are exaggerated, particularly when they come from executives whose companies benefit from promoting the technology.
WGAE executive director Sam Wheeler pointed to worker displacement, degraded online information, rising energy costs, and threats to democratic institutions as harms requiring immediate regulation. He said collective bargaining can protect workers in specific industries but cannot replace government oversight of AI companies.
The widening disagreement shows why enforcing a slowdown remains difficult. Technical mechanisms such as compute monitoring and third-party audits still depend on political legitimacy, international cooperation, and public trust. Without agreement on which risks deserve priority, even executives who support safety measures remain divided over who should set the rules and how those rules should be enforced.
