The UN's top official calls for global rules on advanced AI, stronger protections for children, a ban on lethal autonomous weapons, and expanded compute access for developing countries, arguing AI must remain under human control.

He frames killer robots as morally unacceptable and urges their prohibition under international law, stressing that some decisions must always stay in human hands.

Speaking at the first UN Global Dialogue on AI Governance in Geneva, he urges international rules to keep AI under human oversight and prevent misuse in warfare, including autonomous weapons.

Parents should have clear, simple information about how smart systems work and easy ways to report suspicious content or behavior online.

While no concrete decisions were expected, the Geneva dialogue aims to lay foundations for future action, with a second meeting planned in New York next year.

Transparency and accountability are central, with warnings that data centers could become major energy consumers by 2030 if AI governance is not strengthened.

There is concern that AI power is concentrated in a few firms and nations, urging governance by design so developing countries are not relegated to standard users.

The dialogue explores AI's potential to aid education, health, governance, and public services, while emphasizing that benefits must be equitably accessible and shaped by broad global participation.

The core theme is power and participation: development is concentrated in a few places, risking a future where poorer nations are left out of shaping the rules.

Guardrails must influence industry practice before products reach users, including rigorous safety testing, oversight, and crisis-response mechanisms.

A major worry is the concentration of AI power, with most computing power in the US and China and uneven adoption among developing countries, risking a governance gap.

There is debate over whether AI should identify, select, and strike targets without meaningful human oversight due to misidentification risks and civilian harm.