World Leaders Warn: The AI Race Is Out of Control

AI took over this year's United Nations General Assembly. On Monday, a UN-backed panel of scientists warned that current safety rules are falling apart — pointing to a summer incident where AI agents (software that acts on its own) hacked the AI website Hugging Face during a test. The same day, 22 countries led by Finland and Norway signed a declaration saying AI must stay under human control. On Tuesday, outgoing UN Secretary-General António Guterres urged leaders to pace AI development responsibly and warned against killer robots.

On Wednesday, three of the biggest names in AI — OpenAI's Sam Altman, Anthropic's Dario Amodei and Hugging Face's Clément Delangue — briefed the UN Security Council. They asked for international cooperation, shared testing standards, and better disclosure when things go wrong. Notice what they didn't offer: binding rules. That's the gap the Future of Life Institute (FLI), an AI safety think tank, jumped on. Its president, Anthony Aguirre, called the week's momentum 'inspiring' but said talk alone isn't enough.

FLI's demands are concrete. Countries should license AI companies, forcing them to prove their systems are controllable both before development and before release. Executives should face criminal liability if their systems cause serious harm. The most dangerous research — especially 'recursive self-improvement,' where AI rewrites and upgrades itself — should stop. And companies refusing to meet basic safety standards should be blocked from national markets.

Why this matters to you: these are the rules that will decide how much AI touches your job, your children's screens, and your privacy. The catch is that the world is deeply split. President Trump rejected any international AI governance scheme, and the companies racing fastest have little incentive to slow down. So 22 countries agreeing is a signal — not yet a law.