Editorial illustration for Altman reportedly pitched three AI safety endgames, including Demis Plan
Altman Pitches Three AI Safety Governance Plans
Altman reportedly pitched three AI safety endgames, including Demis Plan
Sam Altman spent part of last week walking OpenAI staff through three possible futures for how the world might govern superintelligent AI, according to a report from The Information citing people familiar with the discussion. The timing isn't incidental. The Trump administration is drafting what would be its first formal framework for overseeing companies building frontier AI systems, and OpenAI wants a position staked out before that framework lands.
The scenarios Altman described range widely in how much power stays distributed versus concentrated. One option borrows from nuclear nonproliferation, building an inspections regime similar to the International Atomic Energy Agency, an idea OpenAI floated publicly back in May 2023. Another draws from a decade-old pitch by Demis Hassabis, now running Google DeepMind, for a CERN-style research body that pools resources across major labs rather than leaving any single company in charge.
The third option is the one getting attention: a world where one firm's compute advantage becomes so large that it can simply outpace, and effectively neutralize, everyone else's hardware. OpenAI reportedly intends to settle on a direction within the next year, all while making sure it doesn't fall behind competitors chasing the same breakthroughs.
Altman reportedly called the third approach the "Eliezer Plan," after AI safety theorist Eliezer Yudkowsky. In this scenario, a developer accumulates so much economic power through self-improving AI that it withholds compute from everyone else and could, in the most extreme case, shut down everyone else's computing capabilities entirely. Altman brought up the scenario only for completeness, according to the report.
Why this matters
What's notable here isn't the menu of options, it's who's holding the pen. Altman is the one sketching the IAEA model, the CERN-style merger, and the single-company scenario, and OpenAI is the one promising to pick a lane within a year while refusing to slow down in the meantime. That's worth sitting with. A company racing for market position is also drafting the rules for how racing should end, and two of the three plans it's floated involve consolidating power in a handful of labs, or one.
For developers and founders building on top of these models, the "Demis Plan" deserves particular attention: a CERN-like merger of major labs sounds like safety-through-cooperation, but it also reads as a blueprint for locking out everyone not already in the room. Researchers should ask who gets a seat at that table and who's deciding the seating chart. Governance proposals from the companies with the most to gain from any given outcome aren't neutral. Watch which plan OpenAI actually moves toward, and who else gets a vote.
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives Meta’s Muse hits 10M U.S. downloads in 40 days, joins 11 other apps
From Archives US Orders Federal Agencies to Remove DeepSeek AI Over Model Distillation
Common Questions Answered
What are the three AI safety endgames that Sam Altman pitched to OpenAI staff?
Sam Altman presented three possible governance models for superintelligent AI to OpenAI staff, according to The Information. These scenarios include an IAEA-style international regulatory model, a CERN-style merger approach for consolidating AI development, and the 'Eliezer Plan' where a single developer accumulates so much economic power through self-improving AI that it could withhold compute from competitors. The timing of this pitch coincides with the Trump administration drafting its first formal framework for overseeing frontier AI companies.
What is the 'Eliezer Plan' scenario described by Altman?
The 'Eliezer Plan,' named after AI safety theorist Eliezer Yudkowsky, describes a scenario where a developer accumulates so much economic power through self-improving AI that it could withhold compute from everyone else and potentially shut down all other computing capabilities. According to the report, Altman presented this scenario only for completeness, suggesting it represents an extreme outcome in the spectrum of possibilities.
Why is the timing of Altman's AI governance pitch significant?
The timing is significant because the Trump administration is currently drafting what would be its first formal framework for overseeing companies building frontier AI systems. OpenAI wants to stake out its position on AI governance before this official framework is released, giving the company influence over how regulatory oversight will be structured.
What concern does the article raise about OpenAI's role in drafting AI governance rules?
The article highlights that OpenAI, a company actively racing for market position in AI development, is simultaneously drafting the rules for how that race should be governed and regulated. This creates a potential conflict of interest, as two of the three governance plans OpenAI has proposed involve consolidating power in a handful of companies, which could benefit OpenAI's competitive position.
What is OpenAI's timeline for committing to one of these AI governance approaches?
According to the report, OpenAI has promised to pick one of the three governance lanes within a year while refusing to slow down its AI development in the meantime. This suggests the company plans to continue advancing its technology even as it publicly debates which regulatory framework should govern the industry.
Further Reading
- Governance of superintelligence - OpenAI
- Altman Proposes Three "Endgame Paths" for AI Development: From International Regulation to "Monopoly" - Longbridge
- Sam Altman outlines 3 AI endgames, including one company dominating all rivals - The News International
- AI takes on great experiment in self-rule - Axios
