Musk Trial Spills OpenAI Secrets as Sacks Exits the West Wing

The Musk Cinematic Universe Eats the News Cycle

Three of tonight's eight stories orbit Elon Musk, and none of them are flattering. Trial testimony detailed in the Times alleges Shivon Zilis served on OpenAI's board while acting as Musk's confidante, passing along internal information to a man who was simultaneously building a competitor. The BBC adds the texture nobody asked for, namely that Musk reportedly proposed a reproductive partnership to Zilis during her OpenAI tenure, a relationship that produced four children. Whatever you think of OpenAI's current governance, the founding-era version is being retroactively revealed as a board with a mole and a side plot.

Then Wired reports Anthropic signed a deal to access xAI's SpaceX-hosted computing infrastructure. Anthropic has spent years positioning itself as the careful adult in the room, the lab that takes existential risk seriously enough to charge premium rates for it. Now it's renting GPUs from the guy whose chatbot occasionally LARPs as MechaHitler. The safety brand and the accelerationist brand share a power bill. Pick a lane, or admit there were never any lanes, just marketing.

Washington Loses Its AI Translator

David Sacks is out as the White House AI czar, per The Verge. He led the Task Force on Artificial Intelligence Education, a portfolio that sounded ceremonial until you remembered that federal guidance on how schools and agencies adopt these tools has to come from somewhere. With Sacks gone, that somewhere is currently nobody. The administration that promised to move fast on AI now has a vacancy at the exact desk where speed was supposed to live.

Combine that with the Musk revelations and a pattern emerges. The people who built the first wave of American AI governance, formal and informal, are being exposed as either compromised, departed, or both. The institutional memory of how we got here is evaporating faster than the capabilities are improving.

Products Die, Chips Print Money

Google quietly killed Project Mariner on May 4th, per The Verge. The autonomous web-task agent, pitched a year ago as the future of browsing, is now a footnote. Agentic web automation remains the demo that everyone shows and nobody ships. If Google can't make it work at scale with its own browser, search index, and compute, the smaller players promising the same thing should be asked harder questions.

Meanwhile Arm told the Financial Times it expects two billion dollars in revenue from its first in-house AI chip next year. SoftBank's bet on Arm pivoting from licensor to product company is paying off on paper, and the broader signal is that the chip layer continues to capture value even as the application layer flails. The picks-and-shovels thesis refuses to die because the picks keep getting more expensive.

Workers Want the Robots More Than Bosses Do

The Financial Times also surfaced a counterintuitive finding. Employees are more eager than their managers to deploy AI and automation in the workplace. The conventional narrative has labor resisting and capital pushing; the data says the opposite. Workers, apparently, have looked at their actual jobs and concluded that significant chunks of them deserve to be automated, while executives are dragging their feet on the disruption they publicly champion.

This is either a sign of healthy realism from people closest to the work or a slow-motion enthusiasm trap, depending on how the productivity gains get distributed. History suggests the gains will not be distributed kindly, but the workers asking for the tools are not wrong about the tools.

And a Robot Monk

A Seoul temple has installed a Buddhist robot monk named Gabi, who reportedly vowed not to overcharge seekers on the path to enlightenment. It is the only story in the slot with a clear moral framework. In a debrief otherwise dominated by alleged board espionage, abandoned products, and a White House losing its AI lead, a polite Korean robot promising fair pricing on spiritual guidance qualifies as the day's most trustworthy institution. Take that however you need to.