Sam Altman Predicts 3 Ultimate AI Outcomes: The Worst-Case Scenario That Could Paralyze All Competitors' GPUs
According to a report by The Information, OpenAI CEO Sam Altman recently described three possible endings for the AI race during an internal company-wide meeting:
The first is to create a global organization to regulate AI. The second is for the leading AI research labs to join together as one organization. The third is for one company to take the lead in making major technological breakthroughs and even gain the ability to disable or destroy its competitors' hardware.
The third possibility sounds like something from a science fiction movie. Altman himself called it an "utterly terrible" outcome.
Three Possible Paths, Each With Its Own Problems
Altman's first proposal is to create a global organization to oversee AI, similar to the International Atomic Energy Agency (IAEA).
This organization could monitor the most advanced AI systems, set common safety rules, and require AI companies to undergo independent reviews.
As early as May 2023, Altman joined OpenAI co-founder Greg Brockman and then Chief Scientist Ilya Sutskever in publishing an article called Governance of Superintelligence.
Link: https://openai.com/index/governance-of-superintelligence/
The article suggests that the world may eventually need an international organization like the IAEA to oversee AI development once systems reach certain levels of ability or computing power. Its work could include inspecting AI systems, checking their safety, testing whether they follow the rules, and tracking their computing power and energy use.
However, AI has become a strategic technology, and countries around the world are investing heavily in it. International rules would raise difficult questions about keeping technology secret, protecting national security, and competing in industry. It remains uncertain whether countries would agree to follow the same rules.
The second path is for the world's leading AI research labs to eventually join together in one organization.
Altman calls this the "Demis Plan." Demis Hassabis once suggested that AI research could follow the example of the European Organization for Nuclear Research (CERN), where several countries work together to build and run a major research organization.
Link: https://www.youtube.com/watch?v=4poqjZlM8Lo
If the leading AI companies were run by the same organization, they could share safety rules and lessons learned. This could help stop companies from constantly rushing to develop AI faster and release products sooner than their rivals.
However, if the most advanced AI models, top talent, and computing power were all controlled by one organization, it would gain enormous technical and economic influence. Preventing too much power from ending up in one place would become a new challenge.
Altman calls the third path, where one company wins and dominates everyone else, the "Eliezer Plan." It is an extreme possibility proposed by AI safety researcher Eliezer Yudkowsky.
Link: https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities
Yudkowsky has long warned that superintelligent AI that escapes human control could threaten humanity's existence. He once argued that whoever first gains control of superintelligence would need to take decisive steps to stop other organizations from developing potentially dangerous superintelligent systems.
Altman connects this possibility to Recursive Self-Improvement (RSI).
RSI means that AI can help improve its own design, training methods, or development process. The improved AI can then keep making further improvements, creating a repeating cycle in which it becomes better and better.
If this process becomes independent enough, AI research and development could speed up greatly.
If one company makes a breakthrough in RSI before its competitors, it could quickly build a technological lead that other companies cannot catch up with. It could eventually control the market.
In the most extreme case, this company might even be able to interfere with or destroy its competitors' GPUs, the computer chips used to run AI systems.
Altman clearly said that he opposes a future in which one company controls advanced AI.
The Risks Altman Has Repeatedly Warned About
This internal discussion continues a conversation about risks that Altman has focused on for some time.
In mid-September 2026, Altman publicly pointed out two major risks from advanced AI. The first is that humans could lose control of AI systems. The second is that AI power could end up in the hands of just a few organizations.
He stressed that steps must be taken to stop AI development from leading to either of these outcomes.
He said OpenAI should act in the interests of all humanity, not just one company or one interest group. Altman also called for stricter AI safety rules and independent checks of advanced AI systems.
This view is similar to some of the views held by Dario Amodei, CEO of Anthropic.
However, Altman also admitted during the internal meeting that OpenAI cannot easily stop competing with other companies to become the leader in AI technology.
Reference link:
This article comes from the WeChat official account "Synced" (ID: almosthuman2014), an AI-focused contributor. It was published with permission from 36Kr.