Sam Altman reveals 3 possible futures of AI: Global watchdog, AI lab merger and a dangerous GPU monopoly

Sam Altman, the CEO of OpenAI, has reportedly talked about three ways artificial intelligence (AI) could develop in the future. One possibility is a worldwide group that checks AI safety. Another is a single organisation that brings together leading AI research companies. The third is a situation where one company controls so much computing power that other companies cannot compete with it.

According to reports, Altman called the third possibility “terrible”. His comments show why people are becoming more worried about AI safety, fair competition, keeping humans in control, and making international rules as AI becomes more powerful.

Artificial intelligence is developing very quickly, but there are still important questions. Who should control it? How can we make sure it is safe? These questions do not yet have clear answers.

Sam Altman, the CEO of OpenAI, discussed three possible ways AI could develop next.

According to The Information, he shared these ideas with OpenAI employees in October 2026.

The first possibility is an international group that watches over advanced AI. The second is a plan for leading AI research companies around the world to work together. The third is a future where one company becomes so powerful that others cannot compete, especially because it controls the computers needed to develop and run AI.

According to the report, Altman seemed to support international supervision of AI. He did not support the idea of one company controlling the whole industry.

He also reportedly said that OpenAI would need to choose which approach to support during the next year.

The first idea is to create an international organisation that checks how powerful AI systems are being developed.

This group would work in a similar way to the International Atomic Energy Agency (IAEA). The IAEA encourages countries to use nuclear energy for peaceful purposes. It also carries out inspections and checks to make sure nuclear materials are being used properly.

Under this plan, an international AI watchdog could create safety rules, inspect powerful AI systems, require independent experts to check them, and make sure companies follow the rules they have agreed to.

It could also help governments understand the dangers of AI systems as they become more capable.

The goal would be to stop companies from building very powerful AI systems without proper safety measures. Countries could work together to create the same basic rules instead of leaving each company to decide for itself what is safe.

This is not a completely new idea. In a policy paper published in 2023, OpenAI's leaders suggested creating an international organisation that could watch over advanced AI development, inspect AI systems, and make sure safety requirements are followed.

Altman's recently reported comments come at a time when governments are discussing how to control AI risks without slowing down useful new technology or giving too much power to a small number of companies.

However, creating such a group would not be easy. Countries would first need to agree on what powers it should have, what it would be allowed to inspect, and how it could make companies follow the rules.

The second idea is called the “Demis plan”. It is connected to an idea from Demis Hassabis, the CEO of Google DeepMind.

The plan involves creating a shared organisation where researchers work together on advanced AI. It is similar to CERN, an international science organisation known for studying particle physics, which looks at the tiny particles that make up matter.

According to the report, Altman's version of this plan would allow leading AI research labs to combine their work or operate through one shared organisation. They could then develop powerful AI systems with greater attention to safety. Countries taking part could get access to AI models created through this shared work.

The main idea is that companies developing powerful AI should cooperate instead of always competing against one another.

By working together, researchers could share what they learn about safety, plan tests together, and avoid making the same mistakes. They might also find it easier to agree on safety measures that everyone follows.

However, there would be some big problems to solve. AI companies compete to attract customers, investors, talented workers, and the powerful computers needed to build AI.

These companies might not want to share valuable research or give up control of their technology.

It is also unclear who would make decisions in such an organisation, how access to AI systems would be shared, or how it would make sure smaller companies and countries are treated fairly. The report does not say that any companies have agreed to merge or create a shared research organisation.

The third idea is the most controversial. The report calls it the “Eliezer plan”, referring to Eliezer Yudkowsky, a thinker who studies the dangers and safety of AI.

In this possible future, one AI company becomes so powerful that it can stop other companies from getting the computing resources they need to build advanced AI systems.

Altman reportedly described a situation where a company could use its financial power to keep important computing resources away from its competitors. These resources include graphics processing units (GPUs). He called this possibility “terrible”.

GPUs are special computer chips that are widely used to train AI systems and help them work.

Building powerful AI systems requires huge amounts of computing power, electricity, data, and special equipment and facilities.

If one company controlled a large share of these resources, other AI developers could struggle to compete. That company could then have a lot of power over which AI technologies get built and who is allowed to use them.

But the worry is about more than just business competition. If one organisation controlled the most powerful AI systems, it could also have a major influence on decisions affecting companies, governments, and ordinary people.

For example, Yudkowsky has warned that creating powerful AI could bring very serious dangers to humanity. However, the situation described in the report is only a possible future. It does not prove that any company today can stop its competitors from getting the computing resources they need.

This discussion is happening as AI systems become better at doing difficult tasks. They can write computer code, carry out research, and do more work without people guiding every step.

These improvements could help make healthcare, education, scientific research, and everyday work better and more efficient.

However, more powerful AI systems could also bring dangers. These include computer security threats, people using AI for harmful purposes, AI making decisions that cannot be trusted, and humans losing control over what AI systems do.

In September 2026, Altman spoke to the United Nations Security Council (UNSC). He called on countries to work together to keep AI safe. He said governments should help create AI rules through democratic processes, where people have a say. He also called for shared standards to measure what AI systems can do, understand their risks, and make sure humans remain in control.

OpenAI has also asked for safety rules that different organisations can follow and better ways to report serious problems involving AI. In its policy proposals, OpenAI says that companies should have their own safety measures, but these should work alongside government supervision rather than replace it.

Altman's reported discussion of three possible AI futures points to an important problem for the technology industry: how to keep building more powerful AI while making sure it stays safe and that the people developing it can be held responsible.

An international watchdog could create common safety rules. A shared research organisation could help AI companies work together. But a situation where one company controls the industry could leave too much power in the hands of just a few people.