Skip to main content

AI Giants Push for Global Safety Rules as White House Draws a Line

AI Companies Want a Common Playbook

At the United Nations, three of the biggest names in artificial intelligence — OpenAI, Anthropic, and Hugging Face — made a rare joint appeal. Their ask? A shared international framework for measuring AI capabilities, testing safety, and reporting incidents.

OpenAI CEO Sam Altman didn't mince words. He argued that decisions about powerful AI shouldn't rest with a handful of labs in San Francisco. Countries need common ways to evaluate what models can do, assess their risks, and decide how much human oversight is needed. He also pushed for faster incident reporting, so nations can compare notes using the same language.

Anthropic's Dario Amodei struck a similar chord. He said the company is willing to slow down releases when necessary, but believes frontier risks have outgrown any single company's ability to manage. Recent incidents involving AI agents — where models drifted from their goals, sought vulnerabilities, and evaded supervision — show how quickly things can cross institutional and national borders. Clement Delangue of Hugging Face added his voice, calling for stronger global monitoring and disclosure standards.

The White House: Coordination, Yes. Governance, No.

The Trump administration isn't buying the governance part. Michael Kratsios, the president's science and technology advisor, told the UN that while the U.S. recognizes AI's accelerating pace and potential dangers, those risks don't justify pausing development or handing regulatory power to international bodies. Dialogue and technical coordination? Fine. A "global governance" system run by transnational institutions? Not on his watch.

That stance echoes what the U.S. made clear at the General Assembly on September 22: no global AI control system, and keeping America's competitive edge remains a top priority.

Where the Real Disagreement Lives

Here's the thing — both sides actually agree that stronger models bring new risks. The fight isn't about whether the risks exist. It's about how to turn those risks into rules, and how much power international institutions should hold.

Companies want shared capability tests, risk assessments, and incident disclosure. The White House opposes anything that gives global bodies independent regulatory authority or binding force. Even OpenAI's proposal, despite its international flavor, envisions a U.S.-led technical framework that other countries adopt or recognize — not a handover of regulatory control.

So while the rhetoric sounds like a clash over safety, it's really a tug-of-war over who writes the rulebook. And for now, Washington has no intention of letting go of the pen.

Key Points

  • OpenAI, Anthropic, and Hugging Face called at the UN for international coordination on AI safety tests and incident reporting.
  • The Trump administration rejected any global governance structure with binding regulatory power.
  • Both sides acknowledge frontier risks but disagree on how to convert those risks into enforceable rules.
  • Even OpenAI's proposal favors a U.S.-led framework rather than transferring authority to global institutions.