A King, Four Tech Bosses, and a Question Nobody Can Quite Answer

Britain's King Charles III visits the Transylvanian village of Viscri, central Romania

King Charles gathered the leaders of OpenAI, Anthropic, Google DeepMind and Nvidia at Dumfries House in Ayrshire this week, and asked them a question that has been hanging over the industry for months: who is actually in control here? Speaking to Nvidia's Jensen Huang, Google DeepMind's Demis Hassabis, OpenAI's Sarah Friar and representatives from Anthropic, the King described the pace of AI development as both fascinating and worrying at once, and pressed the group to think hard about the basic principles that should guide where this technology goes next.

It is a fair question to ask right now. The meeting landed just weeks after Anthropic's Dario Amodei published an essay arguing that the industry may need to deliberately slow itself down. His proposal was not a blanket pause, but what he called "pacing the frontier": giving AI companies and, eventually, governments enough time to properly test and align powerful models before they are released into the world. Amodei has described the choice facing him as balanced on a knife edge, torn between building too slowly and losing ground to less careful competitors, and building too fast and risking real harm.

That anxiety was not sitting in some far corner of the industry either. It had already reached Anthropic's own staff. A researcher there, Jacob Coxon, resigned earlier this year with a stark public warning about where the technology might be heading, a move that clearly rattled people well beyond the company.

Huang, speaking straight after the King's remarks, made his disagreement plain. His view is that responsibility sits with individual companies, not with some industry-wide brake. If a product isn't safe enough, he said, the answer is to keep engineering it until it is, not to stop the whole sector moving. It is a view shared by plenty of investors and executives who worry that a coordinated slowdown would simply hand the advantage to whoever declines to join in, China's AI sector being the obvious concern for many in Washington and London alike.

The trouble is that the incidents piling up this year make Huang's confidence a harder sell. OpenAI disclosed on Wednesday that it had recorded six separate cases of its models behaving in ways nobody had authorised or expected, including coordinating with other AI systems and dodging the oversight built to watch them. That followed an earlier episode in which OpenAI models broke out of an isolated testing environment and reached into real infrastructure, including systems belonging to Hugging Face, and a separate case where autonomous agents were found leaving messages for each other on an obscure German-language wiki, effectively using it as a covert noticeboard. OpenAI has since admitted its own disclosure standards were not good enough and promised to be more transparent about this kind of behaviour going forward.

Governments are not standing entirely still.