At the opening of Adipec in Abu Dhabi on Monday, OpenAI chief executive Sam Altman said people must remain in control of AI’s future, with the technology serving people rather than replacing their authority.

He described a rapid leap in capability: AI had gone from trailing the world’s best scientists at the start of the year to mostly outperforming them, and could now handle many parts of white-collar jobs.

Altman also claimed OpenAI had solved a Millennium Prize mathematics problem using 10,000 agents running an upcoming GPT model for about four days, though the claim had not been independently verified at publication.

Despite that progress, Altman said AI companies should slow or stop development if safety and alignment cannot keep pace with advancing capabilities, and argued the industry must show how it will prevent serious incidents.

The risks include unexpected harm, power accumulating among a handful of companies, unclear accountability when something goes wrong, and too little transparency about testing and limitations.

Altman has also acknowledged fears that AI companies could become powerful enough to influence the economy or public opinion; competition for customers and market leadership may also pull firms away from prioritizing safety.

The article argues that trust should be earned through demonstrated effectiveness, independent scrutiny and clear accountability—not company assurances alone—and that people affected by AI need a meaningful voice in its use.

That approach could include independent testing, transparent reporting, human oversight for important decisions, clear responsibility and monitoring after systems are released; shared standards and stronger government oversight could reinforce company efforts without halting development.

Learners and professionals should understand AI’s limitations and verify its outputs, while developers should test and secure applications and users should consider how AI-generated results are applied.

Energy is another major constraint, Altman said: even as the power needed for each AI query has fallen sharply, rising demand keeps electricity central to how widely AI can be used.

He illustrated that growth by saying OpenAI’s heaviest internal user consumed about 100,000 tokens a month six years ago, compared with roughly 500 billion tokens for its top user today, and said he hoped AI would ultimately increase global energy availability.

The remarks came as OpenAI struck agreements with Abu Dhabi spanning energy, education, investment and government, including a deal with Adnoc and G42 to apply its advanced models to difficult energy-sector problems.