Altman Introduces Three Pathways to Securely Develop Super AI to Employees

Last week, OpenAI CEO Sam Altman laid out three distinct pathways for safely developing superintelligent AI to company employees. The first approach proposes establishing a global regulatory body—modeled after the International Atomic Energy Agency—to review advanced AI models and conduct independent audits of companies’ safety practices; according to insiders, Altman personally leans strongly toward this option. The second draws inspiration from CERN, suggesting that leading AI labs jointly form a single entity—or coordinate R&D under a unified governing body—to replace today’s fragmented, competitive landscape. The third—and most radical—pathway, dubbed the “Eliezer Plan” by Altman, envisions one AI company achieving recursive self-improvement first, then leveraging the resulting overwhelming economic advantage to monopolize compute resources and permanently block competitors’ progress. Altman vividly described this scenario as “melting all other companies’ GPUs,” explicitly labeling it “very bad” and including it solely for completeness—the metaphor originates from discourse within the AI safety community.