AI is on trial

AI is on trial. Elon Musk took the stand Tuesday in an Oakland federal courtroom, seeking roughly $150 billion in damages and asking the jury to force OpenAI back into nonprofit form. He testified he gave about $38 million to OpenAI as a charity, and that Altman and Brockman “stole” it. OpenAI’s for-profit arm is now valued at $852 billion, and its counsel called the case “sour grapes” from a power struggle Musk lost. The verdict won’t just settle a grudge—it will set a legal precedent for how nonprofit AI labs are allowed to convert into private empires.

The corporate version of the same trial is playing out at Goldman Sachs. Reuters confirmed Wednesday that Goldman has blocked Hong Kong bankers from accessing Anthropic’s Claude for several weeks, citing the bank’s own “strict interpretation” of its contract with Anthropic after consulting the company. Gemini and ChatGPT remain available on the same internal platform; Anthropic’s products were the only ones pulled. With a Trump-Xi summit scheduled for mid-May, jurisdictional contract reading is becoming its own form of AI policy.

Then there is the clinical version of the trial—the one AI is winning. A new model called REDMOD, published Tuesday in the journal Gut, flagged pre-clinical pancreatic cancer on CT scans an average of 475 days before clinical diagnosis, with 73% sensitivity versus 39% for experienced radiologists. The authors are explicit about the caveats: the training cohort of 219 patients and 1,243 controls was not ethnically diverse, and prospective validation in high-risk groups is still required. While the courtroom and the bank argue over governance, the imaging model is quietly demonstrating the upside that makes governance worth fighting over.