Close and Friendly

Wednesday in Oakland. Day seven. The trial that started as a billionaire grudge match keeps delivering.

Let's get into it.

Buried in Tuesday's Brockman testimony — and somehow not the headline — was this: Musk enlisted several OpenAI employees to do months of free work on self-driving technology at Tesla's Autopilot team. In 2017. While he was sitting on OpenAI's nonprofit board.

A 2017 email from Musk to a Tesla VP, entered into evidence, reads: "The OpenAI guys are gonna want to kill me. But it had to be done."

He knew it was wrong. He wrote it down. The jury has it.

For context on who this person is: earlier in 2017, right before the meeting where he demanded absolute control of OpenAI, Musk gave each co-founder a Tesla Model 3. Sutskever had commissioned a painting of a Tesla as a friendly gesture for the occasion. When the co-founders refused to hand over control — control he wanted partly to fund what he described as an $80 billion city on Mars — Musk grabbed the painting and stormed out. Brockman testified: "I thought he was going to hit me. He turned around and said 'When will you be departing OpenAI?'"

The man suing OpenAI for self-dealing used OpenAI's employees to do work for his for-profit car company while on the nonprofit board. Under oath. In his own lawsuit.

That was Tuesday. Wednesday brought its own receipts.

Mira Murati was OpenAI's Chief Technology Officer. She was briefly its CEO when the board temporarily fired Altman in November 2023. She left in 2024 and co-founded her own AI startup. She knows where everything is.

Her pre-recorded video deposition played in court Wednesday. The question: did Altman tell the truth when he said OpenAI's legal team had determined a new model didn't need review by the deployment safety board?

Her answer, on the record: "No."

One word. Under oath. From the person who was briefly running the company.

She also testified that Altman was "creating chaos" — saying one thing to one person and the opposite to another. Pitting executives against each other. Undermining her role as CTO. And yet she pressed board members for a fuller explanation when they fired him in 2023 because "OpenAI was at catastrophic risk of falling apart."

The Burn It Down post documented who Altman is before any of this was sworn testimony. Murati just documented it from the inside. Under oath. With one word.

"No."

Then Shivon Zilis took the stand.

Shivon Zilis took the stand Wednesday. The framing in most coverage: mother of four of Musk's children. Former OpenAI board member. Neuralink executive. The person Brockman and Sutskever nicknamed "Proxy Elon" because Musk was so hard to reach that she effectively spoke for him in the room.

They nicknamed her Proxy Elon. Then voted to keep her on the board and trusted her to keep the Elon conflict under control. The distance between those two decisions is where the governance problem lives.

This is the same Elon Musk who has a prediction market with $8 million in trading volume forecasting his weekly post count. Who had his own engineers adjust Twitter's algorithm because someone was trending higher than him. Who banned journalists who irked him. Who owns the platform, controls the algorithm, and still needed the system adjusted in his favor. The board trusted someone to keep him under control.

She described a complicated relationship with Musk. It began with "a one-off" at a corporate off-site event. Later she decided to have children as a single mother and Musk "offered to make a donation" as a platonic sperm donor. And later still, she said, their relationship evolved.

Four children later, they were photographed holding hands at Mar-a-Lago in February. She joined Musk for a meeting with Prime Minister Modi when he was serving in the Trump administration. She is his chief of staff.

All of this while she was sitting on OpenAI's nonprofit board for five years after Musk left.

Asked if it was her job to funnel information to Musk: "Funnel? Absolutely not." Which is also the only possible answer regardless of what actually happened. Her lawyers didn't prepare her to say yes. Even the judge — who has handled a lot in seven days — would have needed a moment for that one. Some admissions require the court reporter to stop and verify the transcript.

Asked about a text she sent saying "the trust game is about to get tricky": "I would have preferred it if I'd written 'trust framework.'"

The jury has the text. They have the context. They have nine people from the Bay Area who have all been in relationships and know that five years of living together means things come up.

Musk, asked during his own testimony last week to describe his relationship with Zilis, started with "She is my chief of staff." Paused. Then clarified the next day: "We live together. And she's the mother of four of my children."

That pause lasted approximately 24 hours. It's my speculation that somewhere Tuesday evening a lawyer had a conversation along the lines of "you know they're going to follow up on that." The clarification Wednesday morning was the controlled version. Make of that what you will.

While Musk watched Zilis testify, his companies had a busy Wednesday.

SpaceX rented the entire Colossus 1 data center — all 220,000 Nvidia GPUs, 300 megawatts — to Anthropic. The same Anthropic that Musk called "woke," "misanthropic," and "evil" in February. After the deal he posted on X: "No one set off my evil detector."

The evil detector. The man whose platform generated 4.4 million inappropriate images in a week has an evil detector. It did not fire on Anthropic.

The reason Colossus 1 was available: xAI's own internal memo revealed Grok was running at 11% model utilization. The industry average is 35-45%. The biggest AI data center in the world couldn't fill itself with its own AI product. So it rented to a competitor. The IPO narrative quietly shifted from "we have the best AI" to "we have the best AI infrastructure."

Also Wednesday: SpaceX filed plans for Terafab — a semiconductor manufacturing facility in Grimes County, Texas. Initial investment: $55 billion. Total potential investment: $119 billion. That exceeds the entire CHIPS Act authorization. Morgan Stanley projects first chip output won't happen until mid-2028 at the earliest. SpaceX acknowledged in its own IPO filing that the orbital data center plan involves "unproven technologies" with uncertain commercial viability.

The House of Cards post said exactly that. The IPO documentation agrees with it.

And then there's this. From last week. Still funny — given his hobbies.

Musk testified he did not know what TL;DR means.

Too Long; Didn't Read. The abbreviation invented by chronically online people. For chronically online people. In his people's natural habitat.

The man who only read the headline of the 2017 term sheet he's suing over. Who posts 25 times a day to 239 million followers. Who has a prediction market on his weekly post volume. Who has redefined what it means to be chronically online.

Didn't know what TL;DR means.

His own AI called him "a danger to everyone around him." Grok knew. He didn't.

Which brings us to something Barry Diller said this week. Outside the courtroom. Worth sitting with.

He trusts Sam Altman. But trust is irrelevant as AGI nears, he said. Not because accountability doesn't matter. Because "the things that are happening are a surprise to the people who are making those things happen." Not "we've got this." Not "Altman's trustworthy so relax." Just: nobody fully knows what's being built — including the people building it. And "once you unleash that, there's no going back."

Resistance is futile. Not because anyone's in control. Because nobody is.

The series has spent months documenting the gap between what these companies say and what their documents show. Murati under oath: "No." Brockman's journal: "it was a lie." Two sets of numbers for banks and investors. CFO frozen out of meetings for asking if the math worked. Secret Tesla work using nonprofit employees.

And the establishment response is essentially: the train is moving.

The judge in Oakland is doing the work. The jury is paying attention. And one of the most prominent voices in American media just said publicly that whether the man building the most consequential technology in history can be trusted is beside the point.

Back in Oakland, one more line from the week worth noting before we close.

Musk testified he felt "like a fool" for continuing to fund OpenAI. Which is the most self-aware thing anyone has said in that courtroom in seven days.

Somewhere between Zilis and the end of the session five minutes were spent arguing about purple boxes in evidence exhibits. Nobody resolved it.

A woman in the gallery was photographed attempting to sleep through testimony. Honestly fair.

Nadella testifies Monday. Altman the week of May 11. Sutskever still to come.

The judge is getting through the day. That's enough.

Read the terms. They're more honest than the marketing.

  #murati