Sam Altman has reportedly sketched out three ways the race to build advanced AI could end. One involves a global regulator, one involves the biggest labs joining forces, and one involves a single company getting so far ahead that it could take out its rivals’ hardware.
That last scenario sounds like the third act of a techno-thriller. Altman, to his credit, called it “terrible.”
The remarks came during an internal discussion with OpenAI employees, according to reports dated October 19-20, 2026.
The three paths
The first scenario is the most familiar. Altman described an international regulatory body modeled on the watchdogs that oversee nuclear technology, one that would supervise AI development and safety.
The second scenario is more unusual. Altman floated the idea of the leading AI labs consolidating into a single entity.
The pitch here is coordination. If the top players operate as one organization, they could share safety standards and practices rather than competing to ship faster than each other.
The third scenario is the one grabbing attention. It centers on a concept called recursive self-improvement, usually shortened to RSI.
RSI describes an AI system that can improve its own design, which then makes it better at improving itself, which makes it better still.
In Altman’s telling, a company that cracks RSI first could gain such a lead that it dominates the market outright. In the most extreme version, that firm could potentially disrupt the GPUs of its competitors. GPUs, or graphics processing units, are the specialized chips that power AI training and deployment.
It was this outcome that Altman labeled “terrible.”
A theme Altman keeps returning to
The internal comments fit a pattern. In mid-September 2026, Altman made public statements identifying two major risks of advanced AI: humans losing control over the systems, and power concentrating in the hands of a small number of organizations.
On September 14, 2026, he argued that AI must not be allowed to lead to either outcome. He framed OpenAI’s position as being on the side of humanity as a whole, rather than any single company or faction.
He also called for stronger safety protocols and independent assessments of AI systems. That put him in step with Anthropic CEO Dario Amodei, whose essay “Pace the Frontier” Altman endorsed in September 2026. The essay advocates for coordinated safety practices and evaluations across the industry.
Concerns about power concentration and AI misalignment have also featured in OpenAI’s public materials for a long time.
Several figures in the AI field, including Elon Musk, have argued that progress should slow down so safety and ethical questions can catch up.
What this means
Each option carries its own problems. An international watchdog would require governments to agree on rules for a technology that many view as a strategic asset. Consolidating the leading labs would solve the coordination problem by creating a different one: a single, enormously powerful organization, which is exactly the kind of concentration Altman has warned against.
There are also reasons for caution in reading too much into the remarks. They were made in an internal setting, and public documentation of Altman’s internal discussions remains limited.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.