Overview
- Altman, in internal remarks reported Tuesday, sketched three high‑level futures for frontier AI: an IAEA‑style international regulator, a CERN‑style merger of major labs, or a single firm winning via recursive self‑improvement.
- He labeled the single‑firm outcome—often called the 'Eliezer plan'—as a worst case in which a self‑improving system could let one company dominate resources and 'melt everyone else’s GPUs.'
- Altman signaled a preference for an international watchdog that would use inspections, audits and tracking of compute and energy to oversee the most powerful models.
- Practical and market barriers make the merger option unlikely because IPO plans, shareholder incentives and competition push labs to stay independent, even as they talk privately about shared safety tests.
- Major developers including OpenAI, Google and Anthropic are working on common testing and evaluation methods, and the shape of U.S. and global rules will affect IPOs, safety enforcement and how quickly those tests become mandatory.