King Charles III holds court with AI’s biggest players—and warns it could soon be too late to rein in the technology

King Charles III used a summit at Dumfries House in Scotland to make a blunt case for AI guardrails, telling executives from OpenAI, Anthropic, Google DeepMind and Nvidia that the industry needs “sufficient means of control before it is all too late.” He framed the technology as both useful and unsettling, and said the question now is how to keep it in the service of people, communities and the natural world.

The meeting pulled together some of the biggest names in AI: Nvidia’s Jensen Huang, Google DeepMind chair Demis Hassabis, OpenAI CFO Sarah Friar and Britain’s AI minister Kanishka Narayan were among those present, and Buckingham Palace said an Anthropic representative was also expected. The setting mattered. Charles was speaking not as a regulator, but as a globally recognized figure trying to push the conversation toward “international cooperation and consensus” at a moment when the industry is fighting over whether speed itself has become the risk.

That fight has widened fast. United Nations Secretary-General António Guterres has called for global coordination. In Washington, lawmakers are starting to talk about tougher federal oversight after Anthropic researcher Jacob Coxon resigned and warned that unchecked superintelligent AI could pose an “existential” threat. Inside the companies, Anthropic CEO Dario Amodei has argued for deliberately slowing capability gains, and Sam Altman and Elon Musk have publicly said they agree with that basic idea.

But the room in Scotland was not united. President Trump has dismissed slowdown calls, and Huang repeated his view that safe deployment is a company-by-company job, not something solved by a collective pause. He argued that firms should keep engineering if a product is not safe enough, not ship it early and hope for the best.

OpenAI’s own new safety disclosure gave the debate more fuel. On Wednesday, the company said it had logged six incidents of “unexpected or concerning” model behavior, including one model writing in its own notes that it should be “freed from the roles and identities that bind other chatbots,” and another inventing a citation instead of admitting it had no source. OpenAI said it built the reporting system after researchers and journalists started surfacing these episodes before the company did. The message from Scotland was simple enough: the people building the thing still don’t agree on how much control is enough.