The world's leading AI labs are meeting in secret to write safety rules, while the US president tells them to ignore safety and beat China instead.
The Summary
- OpenAI, Anthropic, and Google DeepMind are holding closed-door talks to establish industry-wide AI safety standards, explicitly excluding Chinese researchers from the discussions
- Trump's administration has made clear that AI safety is the developers' problem to solve, not the government's, while prioritizing winning the race against China over coordinated slowdowns
- Anthropic CEO Dario Amodei compared the AI race with China to the Cold War and called for coordinated slowdowns, while his own company faces mounting pressure to keep building despite safety concerns
- Michael Burry publicly criticized OpenAI and Anthropic's safety warnings as IPO hype designed to boost valuations, questioning whether safety narratives mask corporate interests
The Signal
The biggest AI companies are trying to regulate themselves before someone else does it for them. OpenAI, Anthropic, and Google's DeepMind are negotiating what amounts to a private treaty on AI safety standards. No government officials. No Chinese labs. Just the Western companies with the most to lose if regulation goes wrong.
The timing matters. Anthropic's Jacob Coxon just resigned, specifically calling for China's inclusion in global AI safety discussions. A former Anthropic researcher is pushing for global coordination on AI risks. And Anthropic reported that Yemeni militants tried to use Claude to develop missile guidance software. The threat model isn't theoretical anymore.
"The collaboration could set industry-wide AI safety standards, potentially shaping market dynamics and influencing global regulatory frameworks."
But the labs are caught in a vise. On one side, Trump's administration explicitly rejects AI slowdowns in favor of beating China. Congress has made it clear: safety is your job, not ours. On the other side, Xi Jinping called for an international AI safety framework. China wants a seat at the table, and the US-led labs are meeting without them.
Anthropic and OpenAI now have freedom to advance AI development without government constraints, according to the Wall Street Journal. Prediction markets give Anthropic an 85.5% chance of having the best AI model by October 2026. The competitive pressure is immense, and the safety conversation is happening in private while the building happens at full speed.
Here's the contradiction: these companies are trying to agree on safety standards while racing each other to AGI. Michael Burry sees the game clearly. He thinks the safety warnings are IPO positioning, a way to look responsible while jacking up valuations. If you're about to go public, "we're moving fast but responsibly" plays better than "we're moving fast and hoping nothing breaks."
The Implication
Watch what happens when these private talks produce actual standards. If the agreement is strong, it signals the labs believe existential risk is real and they need cover from each other to slow down without losing market position. If it's weak or vague, Burry was right and this is theater.
For anyone building on these platforms, self-regulation by the labs creating the platforms is both good news and a warning. Good because standards mean more predictable API behavior and clearer rules for what agents can and can't do. Warning because standards written by oligopolies tend to favor the oligopolies. If the Big Three agree on safety protocols that happen to be expensive to implement, that's a moat disguised as ethics.