Is Big Tech’s AI Slowdown a Safety Pact or a Cartel?

Summary

OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Google DeepMind cofounder Demis Hassabis, and SpaceX head Elon Musk loosely agreed to slow AI development while pursuing a proposal to “pace the frontier.” The three-part proposal associated with Amodei calls for third-party auditors inside AI labs, regulation of domestic frontier companies, and an eventual global slowdown agreement. Supporters say these measures reflect demands that AI safety advocates and more than 1,000 lab employees have made for years, while critics suspect the companies are trying to delay competitors, restrict open source development, or replace binding law with self-regulation. Industry and safety experts described the commitment as potentially valuable but emphasized that it remains verbal and needs measurable, enforceable requirements. Suggestions include giving auditors access to labs’ compute budgets and requiring significant reductions in research compute; external groups such as METR, Apollo Research, and Redwood Research were cited as possible auditors. The article says voluntary frameworks could become safety-washing or regulatory capture, especially when companies define the safeguards themselves, and some experts compared the idea with weak corporate oversight efforts in social media. Government regulation is viewed as unlikely under the Trump administration, apart from model pre-release review periods already accepted by some labs. China is a central obstacle: US leaders often argue that slowing would surrender an advantage, while Chinese officials have criticized slowdown calls as fearmongering. Some researchers counter that US coordination remains worthwhile and that international cooperation is possible even if China does not immediately participate. The underlying concern is recursive self-improvement, in which AI systems could train and improve successors with limited human involvement. Anthropic has said this milestone could arrive as early as 2027, and researchers across frontier labs warn it could outpace society’s ability to understand or control systems and intensify cybersecurity and catastrophic-risk concerns. The debate therefore turns on whether AI companies can accept credible limits, or whether only independent public oversight can make a slowdown real.