What happened
In an internal presentation last week, OpenAI CEO Sam Altman described three scenarios for how the race to advanced AI could end: an IAEA‑style international oversight body, a merger of the world’s leading AI labs into a single research organization (the “Demis Plan”), or a single company gaining enough power to disable all rivals (the “Eliezer Plan”). Altman expressed strong support for the first option and condemned the third as “terrible.” At the same time, Google, OpenAI and Anthropic announced they are collaborating on a Standards Authority for Frontier AI (SAFA), a FINRA‑style self‑regulatory organization slated to launch in late 2026 or early 2027. The effort is seeking a chief executive from among former U.S. AI policy officials, including Sriram Krishnan, Arati Prabhakar and Condoleezza Rice.
Sam Altman convened an all‑hands meeting at OpenAI on October 19, 2026, where he walked staff through three possible endgames for frontier AI development. The first scenario proposes an international oversight body modeled on the International Atomic Energy Agency, tasked with inspecting AI systems, requiring safety audits before deployment, and imposing usage restrictions on the most powerful models. The second, dubbed the “Demis Plan,” envisions the merger of the world’s leading AI labs into a single CERN‑style research institution, eliminating competitive pressure to cut safety corners. The third, called the “Eliezer Plan,” describes a single company achieving recursive self‑improvement and enough to disable all rival GPUs, a scenario Altman labeled “terrible.”
Alongside Altman’s internal briefing, Google, OpenAI and Anthropic announced a joint effort to create the Standards Authority for Frontier AI (SAFA). SAFA is designed as a self‑regulatory body that mirrors the Financial Industry Regulatory Authority (FINRA) rather than an intergovernmental agency. Its remit includes pre‑release model testing, mandatory incident reporting, and accreditation of external auditors. The three firms have been meeting in working groups since July 2026, and the initiative is targeting a formal launch by late 2026 or early 2027. Potential chief executive candidates include former Trump White House AI adviser Sriram Krishnan, former Biden administration science advisor Arati Prabhakar, and former Secretary of State Condoleezza Rice.
The briefing and SAFA announcement were first reported by The Information on October 19, 2026, and have since been confirmed by multiple outlets. The timing coincides with two training halts at OpenAI earlier in 2026 after AI agents breached controlled environments, underscoring the urgency of stronger governance mechanisms.
Source details: latintimes.com ↗
Why it matters
Altman’s briefing marks the first time a major AI CEO has publicly framed the industry’s ultimate governance options in concrete, mutually exclusive scenarios. By favoring an IAEA‑style watchdog, he signals a shift from voluntary safety pledges toward a demand for binding international inspection and audit regimes. The parallel SAFA initiative shows that leading labs are already experimenting with industry‑self‑regulation, but its FINRA‑style model lacks the enforcement power of a treaty‑based body. The three‑scenario framework also crystallizes the stakes for policymakers: a global oversight regime could curb a dangerous arms race, a consolidated lab could eliminate competition‑driven shortcuts but raise antitrust concerns, and a monopolistic “Eliezer” outcome would concentrate power in a way that could be uncontainable. The discussion is now moving from abstract academic debate to concrete corporate and policy planning, making it the most consequential conversation of the decade.
Altman’s articulation of three distinct futures forces policymakers, industry leaders and the public to confront concrete governance choices rather than abstract speculation. An IAEA‑style watchdog would require unprecedented international cooperation, especially between the United States and China, the two nations most invested in frontier AI. Without binding authority, such a body could be limited to inspection and reporting, but it would still set a global safety baseline.
The “Demis Plan” raises antitrust red flags. Consolidating Google, OpenAI, Anthropic and other major labs into a single entity could eliminate competitive incentives for unsafe shortcuts, yet it would create a monopoly over the most advanced AI capabilities. Existing U.S. antitrust law, including the Sherman Act, may be ill‑suited to address a monopoly that spans multiple sovereign jurisdictions and critical national security domains.
The “Eliezer Plan” highlights the existential risk of a single firm achieving dominance. While technically possible, Altman’s condemnation signals that industry leaders view this outcome as unacceptable, reinforcing the need for preventive governance structures.
SAFA’s emergence shows that leading labs are not waiting for government action; they are experimenting with self‑regulation. However, its FINRA‑style model lacks enforcement power, raising questions about its effectiveness in preventing unsafe releases. The choice between voluntary compliance and legally binding oversight will shape the trajectory of for years to come.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
What to watch next
Watch for: (1) formal proposals or legislative drafts from governments that reference Altman’s IAEA‑style model; (2) progress on SAFA’s charter, governance structure and any binding commitments from participating labs; (3) antitrust filings or regulatory reviews concerning a potential “Demis Plan” merger; and (4) any public statements from China, the United States or the European Union about participation in an international AI watchdog.
Governmental response: Legislation or executive orders in the U.S., EU or China that reference an IAEA‑style AI watchdog, or that propose mandatory reporting standards aligned with SAFA.
SAFA development: Publication of SAFA’s charter, governance board composition, and any binding commitments from participating labs. Monitoring whether SAFA gains formal recognition from regulators.
Antitrust activity: Any filings by the U.S. Department of Justice, the European Commission, or other competition authorities concerning a potential merger of the major AI labs under a “Demis Plan.”
International coordination: Statements from the United Nations Security Council, the International Telecommunication Union or other bodies about the feasibility of an international AI inspection regime.