OpenAI CEO Sam Altman has proposed three potential strategies for global AI safety, which the company aims to evaluate by the end of next year. The first approach involves creating an international regulatory body similar to the IAEA, which would establish safety standards and conduct independent audits of high-performance AI models before their release.
The second option, dubbed the 'Demis plan' after Google DeepMind co-founder Demis Hassabis, suggests forming a collaborative research organization modeled on CERN. This structure would allow developers to share safety research and model assessments, though Altman noted that the required level of international government cooperation makes this difficult to implement.
The third and most controversial scenario involves a company that achieves AGI using its market dominance to restrict competitors' access to GPUs. Altman described this as a terrible option, presented only for the sake of discussion. Some employees speculated that this proposal was directed at the competitor Anthropic.
Meanwhile, OpenAI is in talks with Anthropic and Google to establish a unified oversight body for AI development standards. The parties aim to implement procedures by late 2026 or early 2027, potentially placing independent auditors within each company to monitor safety reports and incident disclosures.
