Musk Tells the Court OpenAI's IPO Puts Humanity at Stake

01On the witness stand, Musk made OpenAI's IPO trial about saving humanity

Elon Musk took the stand in a Northern California courtroom this week. He opened not with OpenAI, not with Sam Altman, not with the contracts at issue, but with South Africa. The jury heard about his arrival in Canada with 2,500 Canadian dollars in traveler's checks. They heard the rest of his biography in chronological order. Then he told them what the case was really about: he just wants to save humanity.

The framing was deliberate. Musk is the plaintiff, but the company in the dock is OpenAI itself. The case, brought against Altman and the organization Musk co-founded in 2015, asks the court to decide whether OpenAI is permitted to operate as a for-profit enterprise at all. The two have been in a yearslong legal feud, according to MIT Technology Review, and a ruling could go further and force Altman out of the company he runs.

OpenAI began as a nonprofit research lab. It has since restructured into a capped-profit hybrid and is preparing for one of the most anticipated IPOs in the private market. Musk alleges the conversion betrays the charitable mission he funded. OpenAI says the structure is legal and necessary to compete. The jury is being asked, in effect, to rule on the legitimacy of that pivot, with the IPO calendar as backdrop.

His opening, then, was less a statement of fact than a positioning move. By the time he finished, the jury had a frame. He was an immigrant who arrived with a few thousand dollars and went on to build rockets and electric cars. Now he stood in court because the AI lab he co-founded had, by his account, abandoned its mission. Altman and OpenAI's lawyers will spend the coming days trying to reframe the same facts as an ordinary corporate dispute over equity and influence.

A ruling for Musk narrows OpenAI's path to public markets and could force the for-profit subsidiary that houses its commercial business to be unwound or restructured. Going the other way leaves the conversion intact and the IPO clock running. The trial is expected to run through the week.

02Three Hacker News threads last week repriced what an AI coding subscription buys you

Within the same week, three of the most-debated developer posts on Hacker News circled the same question from different angles: what exactly does an AI coding subscription buy. Cost, ownership, and autonomy each came up for debate. The three threads were not about new model capabilities. They were about the contract.

GitHub announced Copilot is moving to usage-based billing. The shift means workloads that were predictable under a flat monthly fee now meter against actual consumption. For teams running Copilot at scale, the math that justified seat licenses a year ago no longer holds. Premium request quotas, overage charges, and per-token costs now sit between the developer and the model.

A Substack post titled "Who owns the code Claude Code wrote?" drove the second thread. The piece argued that terms of service have not cleanly answered who carries the copyright when an agent generates substantial portions of a codebase. It also flagged that no clear indemnity applies if training data surfaces verbatim in generated output. Most current developer agreements, Anthropic's included, leave the user with the legal exposure.

The thread that drew the largest crowd, at 848 points, came from an independent developer arguing that AI should elevate thinking rather than replace it. Productivity gains from agent-driven coding can mask skill atrophy, the post warned. It outranked the billing news on the front page. Developers a year into these workflows are asking whether they have outsourced the work that used to make them better.

Read together, the three threads repriced the AI coding subscription. The flat $10 or $20 monthly seat that defined 2024 assumed predictable cost, clean IP, and friction-free adoption. Each default is now contested at once. The next year of vendor pitches cannot rest on any of them; procurement, legal, and engineering all get a say in what gets bought.

03Pentagon Pulls Google In, Brussels Pries It Open, Beijing Locks Meta Out

Governments in three capitals reached the same US AI companies last week and pulled them in opposite directions.

Brussels moved first. The European Commission notified Google that Gemini's default placement on Android likely violates the Digital Markets Act. Rivals including ChatGPT and Perplexity must get equal access to the assistant slot, regulators argued. Google called the demand "unwarranted intervention" and said it would contest. The same DMA logic that already forced browser-choice screens is reaching into on-device AI.

Washington went the other way. The Pentagon and Google reportedly signed a classified contract permitting "any lawful" use of Google's AI, according to The Verge. Earlier defense agreements carved out specific applications. This one removes the categorical fence Google's staff demanded after the 2018 Project Maven walkout. Internal objections did not block the signing.

Beijing closed the third door. China's market regulator rejected Meta's proposed acquisition of Manus, the Singapore-based AI agent startup whose research team is largely Chinese. CNBC reported the block came on national security grounds tied to cross-border data and personnel. Manus had been one of the rare Chinese-founded AI labs courting a US acquirer. That exit path is now closed.

The three actions share a target but pull apart. Washington is binding frontier-AI firms tighter into the defense apparatus. The European push runs the other way: dilute Android's default power. China has sealed AI talent and IP behind its border. Each move ran through a different instrument: a defense procurement contract, an antitrust statute, a foreign-investment review.

Google now has to argue in Brussels that the same product the Pentagon just bought wholesale should not be required to share the Android home screen. Whatever the Commission decides will set the precedent for how every other US AI company defaults itself onto European hardware.