Sam Altman, CEO of OpenAI and creator of ChatGPT, warns of a potentially dangerous AI race, marked by fierce competition and uncontrolled existential risks.

Context of Altman’s statements

Sam Altman expressed his concerns in interviews and events in 2026, highlighting an uncontrolled acceleration in AI development. He describes a “tight race” where OpenAI must prioritize enterprise in 2026 against Anthropic and Google, while investing $1.4 trillion in infrastructure. This warning comes after an internal “code red” against Google’s Gemini, though Altman downplays the manipulability of benchmarks.

The Risks of a Runaway Race

Altman warns against an escalation where speed takes precedence over safety: massive energy consumption by AI (compared to that of humans), “AI washing” (overselling inefficient AI), and job displacement via autonomous agents. He predicts a shift from the investment phase to the “harvest” phase in 2025-2026, with job losses despite productivity gains. China is progressing “astonishingly fast,” intensifying the geopolitics of AI.

What Measures Does Sam Altman Propose to Regulate the AI Race?

Sam Altman, CEO of OpenAI, proposes concrete measures to regulate the runaway race in AI, notably at the AI Impact Summit in New Delhi in February 2026.

Key Proposal: A Global Agency Like the IAEA

Altman calls for the creation of an international agency modeled on the International Atomic Energy Agency (IAEA) to oversee advances in AI. This entity will ensure global coordination, inspect critical developments, and react quickly to risks such as biosecurity (e.g., the creation of pathogens via open-source models).

Urgent Regulation and Protections

He insists on an “urgent” need for binding rules for any technology this powerful, including standards on safety, transparency, and the non-proliferation of dangerous models. In contrast to risky centralization, he advocates multilateral regulation to prevent domination by any one country or company.

Democratization vs. Concentration

For Altman, the key is to democratize AI: sharing the benefits through open governance rather than concentrating technological power, which “would lead to ruin.” He rejects public guarantees for its infrastructure but encourages states to build their own sovereign data centers.