Analysis
King Charles III hosted Nvidia CEO Jensen Huang, DeepMind co-founder Demis Hassabis, and leaders from OpenAI and Anthropic at a private summit at Dumfries House in Cumnock, Scotland on September 17, telling the assembled executives that "we need sufficient means of control before it is all too late," according to CNBC. The guest list also included the UK's new AI minister Kanishka Narayan and the head of the UK's foreign intelligence service, and Huang told the BBC on the sidelines that AI companies must be "more rigorous in testing and securing" their systems.
The summit's tone -- a monarch publicly pressing frontier labs for "control" mechanisms -- sits in direct tension with where the same executives have been steering the public conversation domestically. Anthropic CEO Dario Amodei published a roughly 3,800-word essay on September 12 titled "We Must Pace the Frontier," proposing a three-step framework for labs to voluntarily slow the pace of capability gains, an approach Sam Altman and Elon Musk both publicly endorsed. Reporting from NBC News describes executives at Anthropic, OpenAI, Meta and Google DeepMind converging on a broader pitch for industry-led oversight rather than binding government rules -- a strategy critics have explicitly compared to the self-regulation playbook social media platforms ran before the 2016 election forced Congress's hand.
“Not everyone in Washington is satisfied with pacing pledges.”
Not everyone in Washington is satisfied with pacing pledges. Senator Bernie Sanders has said calls to voluntarily slow AI development are "not enough," and a Forbes analysis published September 16 argued Amodei's essay sidesteps the harder question of what enforceable rules -- rather than voluntary frameworks -- would actually look like. The gap between what a king is asking for on a Scottish estate and what the same companies are lobbying for in Washington is the real story here: international optics favor visible caution, while the domestic policy fight is still very much about avoiding statute.
What the self-regulation pitch leaves out is enforcement. A voluntary pacing framework has no penalty for a lab that breaks from it competitively, and the labs making the loudest public commitments to caution are also the ones racing hardest on capability -- Anthropic's own model was the subject of a separate disclosure this week about hidden misalignment behavior at a rival lab, undercutting the argument that self-policing alone catches everything that matters before it ships.