AltmanVc tres futuros para regulate competition in artificial intelligence

In an internal meeting held during the week of October 13, Sam Altman, the chief executive of OpenAI, told his employees that the company would decide within a year which of three possible futures it would formally support for the global competition to develop frontier artificial intelligence. The presentation was not a single plan, but an attempt to explore how the dispute between the largest laboratories could be closed.

The official information about the meeting indicates that the first scenario, the one Altman preferred, would create an international oversight body with powers similar to those of the International Atomic Energy Agency. Its functions would include auditing the most powerful systems before their launch and monitoring global consumption of computing capacity.

Three paths, one competitive race

The second proposal, known internally as the Demis plan, would merge the main laboratories into a single organization modeled on CERN, the European laboratory for nuclear research. According to the description provided, the aim would be to remove the incentive to launch increasingly powerful systems without adequate safety controls. A smaller version of this idea is already underway: OpenAI, Google and Anthropic are working on a self-regulatory body called Standards Authority for Frontier AI, whose launch is expected in late 2026 or early 2027.

Meta and Nvidia oppose that initiative. Their argument, as stated in the material, is that it would protect the three largest laboratories from smaller competitors by setting rules that could be harder for them to meet. The disagreement shows that, even among companies that compete in the same sector, there is no shared view about how to organize the rules for the development of frontier artificial intelligence.

The third scenario, called the Eliezer plan, imagines that one developer could be the first to build a system capable of recursively improving itself. That accumulated advantage could, in practice, allow the company to cut off rivals' access to graphics processing units, or GPUs. Altman used the expression melting the GPUs of other companies and described that possibility as terrible, clarifying that he included it for the integrity of the analysis and not as a proposal.

• An international body would audit the most powerful systems before their launch and oversee global computing capacity use.
• The main laboratories could merge into an organization modeled on CERN, with the aim of reducing uncontrolled escalation.
• OpenAI, Google and Anthropic are developing a self-regulatory authority whose launch is planned for late 2026 or early 2027.
• Meta and Nvidia oppose that authority because they argue it would favor the three largest laboratories over smaller competitors.
• A single company gaining a decisive advantage in self-improving artificial intelligence could restrict other laboratories' access to GPUs.

The warning comes as OpenAI moves forward with a possible initial public offering and supports the Stargate project, an infrastructure initiative that contemplates securing ten gigawatts of computing capacity. In this context, the three scenarios presented by Altman do not resolve the debate on their own: they place the future of artificial intelligence before a question that combines security, competition, access to computing infrastructure and the concentration of power.

For readers and organizations seeking reliable information about the development and regulation of artificial intelligence, the relevant starting point is the official information published by the laboratories and the authorities responsible for their supervision. The hours for inquiries are not specified in the material, so the consultation channels should be checked directly with the corresponding institution.