OpenAI CEO Sam Altman has outlined three potential frameworks for AI safety to his employees. These proposals were discussed internally last week as the US government prepares its first regulatory framework for superintelligence developers.
The first approach involves establishing an organization similar to the International Atomic Energy Agency. This body would monitor energy and compute usage, conduct system inspections, and enforce safety standards through international cooperation, including with China.
The second option, dubbed the Demis Plan, suggests merging major AI laboratories into a single entity or operating under a unified umbrella. This organization would then distribute models to cooperating nations, mirroring the structure of CERN.
The third scenario, referred to as the Eliezer Plan, describes a situation where a single developer gains enough economic power to restrict compute access for others or potentially disable their computing capabilities entirely. Altman reportedly presented this as a theoretical extreme.
OpenAI is currently balancing these safety discussions with competitive pressures, as both OpenAI and Anthropic are reportedly preparing for potential initial public offerings. OpenAI aims to determine its strategic direction within the next year.
Industry leaders, including Meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang, have expressed concerns that proposed safety regulations could protect incumbent companies from competition. Meanwhile, OpenAI, Google, and Anthropic continue to work on the Standards Authority for Frontier AI.
