OpenAI CEO Sam Altman outlines three endgames for AI development
Altman reportedly told employees that advanced AI could develop in three main ways: a global watchdog could oversee it, the leading AI labs could join together, or one company could get so far ahead that its competitors could not keep up.
FoxTPNL / Wikimedia Commons (CC BY 4.0)
Sam Altman has reportedly described three possible ways the race to build advanced AI could end. The first involves a global regulator. The second involves the biggest AI labs joining together. The third involves one company getting so far ahead that it could even damage its rivals' computer hardware.
That last possibility sounds like something from a technology thriller. Altman himself called it “terrible.”
The comments were reportedly made during a private discussion with OpenAI employees, according to reports published on October 19–20, 2026.
The three possible paths
The first possibility is the most familiar. Altman described an international organization that would oversee AI development and safety. It would work like the groups that monitor nuclear technology around the world.
The second possibility is less common. Altman suggested that the leading AI labs could combine to form one organization.
The idea is to help them work together. If the biggest AI companies operated as one organization, they could share safety rules and methods instead of racing against one another to release new products faster.
The third possibility is the one attracting the most attention. It involves an idea called recursive self-improvement, usually shortened to RSI.
RSI means that an AI system can improve how it is built. Those improvements could help it improve itself even more, leading to further improvements over and over again.
According to Altman's explanation, a company that achieves RSI before its competitors could gain such a big advantage that it might come to control the market. In the most extreme case, that company could even interfere with the GPUs used by its competitors. GPUs, or graphics processing units, are special computer chips used to train and run AI systems.
Altman called this possible outcome “terrible.”
An issue Altman often talks about
The reported comments match concerns Altman has raised before. In mid-September 2026, he publicly identified two major risks from advanced AI: people losing control of AI systems and power becoming concentrated in the hands of a small number of organizations.
On September 14, 2026, he argued that AI must not lead to either of these outcomes. He said OpenAI should work for the benefit of all humanity, rather than serving only one company or group.
He also called for stronger safety rules and independent checks of AI systems. This put him in agreement with Anthropic CEO Dario Amodei. In September 2026, Altman supported Amodei's essay, “Pace the Frontier.” The essay calls for AI companies to work together on safety practices and ways to test their systems.
OpenAI has also discussed the risks of concentrated power and AI systems acting in ways that do not match human goals in its public materials for many years.
Several people in the AI field, including Elon Musk, have argued that AI development should slow down so that safety and ethical questions can receive enough attention.
What these possibilities mean
Each option has its own problems. Creating an international watchdog would require governments to agree on rules for a technology that many consider important to national security and economic power. Bringing the leading AI labs together could make cooperation easier, but it would also create another problem: one extremely powerful organization. This is the same kind of concentration of power that Altman has warned about.
It is also important not to read too much into the reported comments. They were made during a private discussion, and there is still limited public information about what Altman said in these internal conversations.