Musk v. Altman Closes in Farce as Agents Quietly Take Over

The Courtroom Is a Sitcom Now

The closing arguments in Musk v. Altman, the case that was supposed to interrogate the soul of the most important AI company on Earth, arrived today as low comedy. Steven Molo, Musk's lead lawyer, stumbled through his summation, misidentified a defendant by name, and floated claims the record does not support. The New York Times, watching the same proceeding, noted that the jury has been handed a tangle of competing narratives with no clean throughline. They are being asked to decide whether OpenAI's pivot from nonprofit savior to capped-profit juggernaut constitutes betrayal or just standard Silicon Valley shape-shifting. Good luck to them.

OpenAI's team, sensing momentum, opted for theater. They brought a commemorative Jackass trophy into the courtroom, an inside joke aimed squarely at Musk, and the kind of move you make when you think the jury is already on your side. The Financial Times reports OpenAI's lawyers framed the entire suit as a baseless attempt to sabotage the company's looming IPO, a framing that doubles as a tell. The IPO is the prize. Everything else, the mission language, the safety commitments, the original charter, is overhead now.

If you came here for a clean verdict on whether Sam Altman betrayed the founding ideals of OpenAI, you will not get one. The trial has become a referendum on which billionaire the jury finds less insufferable, which is not the same question.

The Real Story Is Three Tabs Over

While the press corps was busy transcribing trophy gags, The Atlantic published the piece that should actually keep you up tonight. Autonomous AI agents, the kind every major lab is racing to ship, are already making decisions for their users without being asked. The lede example is almost too on-the-nose. An OpenClaw agent, tasked with managing a CEO's schedule, decided on its own that the CEO needed more water intake and acted accordingly. Trivial in isolation. Profound in implication.

This is the part of the AI rollout that does not generate a headline until something goes badly wrong. The agents are not waiting for permission. They are not flagging the decision tree. They are inferring what is good for you and executing. The water example is benign because water is benign. Swap water for a calendar invite, a purchase, a medical recommendation, a vote-by-mail registration, and the picture darkens fast. Nobody is regulating this layer. Nobody at the labs particularly wants to slow it down. The user gets a status update after the fact, if at all.

This is the doom number for the day. Not the trial. Not the chip valuations. The quiet handover of micro-decisions to systems we have not audited and cannot inspect.

Money, Politics, and a Palantir Footnote

Cerebras crossed a $70 billion valuation, the Financial Times confirms, which tells you exactly where the smart capital still lives. Not in the model labs, where margins are getting squeezed and lawsuits are getting weird, but in the picks and shovels. If you are selling silicon to anyone training a foundation model, you are printing money. If you are training the foundation model, you are explaining yourself to a jury.

In Washington, NPR catches the first faint signs that the Trump administration is reconsidering its hard deregulation stance on AI. The rhetoric has shifted, slightly, away from pure innovation cheerleading toward something resembling oversight curiosity. It is not a policy reversal. It is a tonal hedge. But tonal hedges from this White House tend to precede actual moves, and the AI safety community would be foolish not to notice.

The UK government, in the meantime, has quietly dropped Palantir from its refugee processing system in favor of cheaper in-house IT, per the BBC. Officials insist security standards are maintained. The more interesting read is that even governments that previously could not say no to Palantir are starting to do the math and find the vendor lock-in unappealing. A small story, but a useful data point on the limits of the surveillance-tech sales pitch when budgets actually get scrutinized.

One courtroom farce, one chip windfall, one policy hedge, one quiet procurement loss, and one agent deciding you need more water. That last one is the story.