I sat down with her to push on the hard parts: why the bill leans on agency power at the exact moment the Trump administration is stripping it and suing states over their AI laws; what it means that a $140 million super PAC backed by Andreessen Horowitz, OpenAI’s president, and a Palantir co-founder is spending to beat candidates who back regulation; and why she keeps comparing AI to the early days of commercial aviation.

We get into the evidentiary trap that makes algorithmic discrimination so hard to prove, her four-level map for governing AI, her fear of an AI “Fukushima,” and her blunt fallback if Congress does nothing: the states.

While New York was asleep, traders in Seoul were fighting for their lives last night. South Korea’s KOSPI fell about 10 percent, led down by Samsung and SK Hynix, the two firms that make most of the world’s memory chips. By the time the opening bell rang here, the contagion had crossed to our shores. The Nasdaq dropped around two and a half percent, and a trillion dollars of American market value bled out.

Why should a bad day on a Korean exchange cost Nvidia anything here? Because the whole AI industry balances on a small number of physical chokepoints, and Seoul holds one of them. Memory chips come from Korea. The advanced logic — chip jargon for the processors that do the computing — comes from a single company in Taiwan. The rare-earth elements get refined almost entirely in China. Three weeks ago the Nasdaq fell 4.18 percent in a single day, its worst since April 2025, seemingly on the same nerves. Investors are taking a new look at the AI industry and asking what exactly their money buys.

Fair question, because on paper, the industry doesn’t make much sense. A widely cited 2025 MIT study found that something like 95 percent of organizations pouring cash into generative AI have gotten nothing back. Five companies now account for close to a third of the entire S&P 500, a concentration the market hasn’t seen in half a century. And there’s another wall looming.

A June report from the Carnegie Endowment added up every nuclear deal the big AI companies have signed, assumed all of it gets built on schedule, and found it would still cover less than a fifth of the power the industry expects to need by 2035. So how on earth will all this pencil out? And if the bet comes up short, whose lights get more expensive?

Read more

A.I. Disclosure: I use LLM technology to help with research, fact-checking, document summaries, editing, and rewrites. I’m trying to use it responsibly, but I’m learning as I go. You can read my full ethics disclosure here.

For seven years the record for the largest stock offering ever belonged to a national oil company owned by a royal family. On Friday, SpaceX nearly tripled it — $75 billion raised at a $1.75 trillion valuation. The all-time list now reads, top to bottom, like a short history of what the public has been tricked into funding. It doesn’t suggest that the markets have much wisdom to offer.

Here are the 10 biggest IPOs ever, and what each one was actually selling.

1. SpaceX (2026) — $75 billion. A company that loses billions a year, priced on a future that hasn’t arrived, with one man holding 82 percent of the vote. What that structure actually is, and the 1602 company it descends from, is in the companion piece I published this morning.

2. Saudi Aramco (2019) — $25.6 billion. A government selling a sliver of the family business. Saudi Arabia floated about 1.5 percent of its state oil company, and the shares went mostly to Saudi citizens and Gulf sovereign-wealth funds rather than to the international investors the roadshow had courted. The offering wasn’t really a bid for capital. It was a bid for legitimacy — an IPO run as an instrument of statecraft, with the kingdom valuing its own crown jewel at $1.7 trillion and daring the market to disagree.

3. Alibaba (2014) — $21.8 billion. The largest U.S. listing of its era sold something the buyers never actually owned. Chinese law barred foreigners from holding the operating company, so investors on the New York Stock Exchange bought stock in a shell company in the Cayman Islands that held contractual claims to Alibaba’s profits. The market funded a legal workaround and called it equity. Talk about vaporware: investors bought the exposure and never the thing itself.

4. SoftBank Corp (2018) — $21.3 billion. Japan’s biggest offering was a debt transfer dressed up as growth. The parent company, loaded with borrowings from its tech bets, spun its mobile unit onto the public to pay down what it owed. The stock fell about 14.5 percent on its first day — a reminder (ahem) that a blockbuster opening and a sound investment are not the same event.

5. NTT DoCoMo (1998) — $18.1 billion. A mobile carrier floated at the tail end of Japan’s stock bubble, the same bubble that in 1990 had briefly made its parent, NTT, the first company in the world worth more than $100 billion. The valuation was a monument to a moment, not to durable earnings, and the hangover lasted a decade.

6. Visa (2008) — $17.9 billion. Visa went public six months before Lehman Brothers collapsed, and it thrived through the wreckage anyway. It doesn’t take a market risk; it takes a cut of every transaction in any weather. Among a list crowded with bets on the future, it’s the rare one that sold the public shares in a reliably profitable business.

7. AIA Group (2010) — $17.8 billion. The Asian life-insurance arm of AIG, carved out and sold to the public to help repay the U.S. government bailout that had kept AIG alive after the 2008 crash. Taxpayers covered the rescue; then the public was invited to buy the salvage. The cleanup of the crisis was in a weird way the product being sold? I know, I don’t get it either.

8. Enel (1999) — $16.5 billion. Italy floated its state electricity company, and citizens lined up to buy shares of a utility their own government had built and owned on their behalf. Privatization, in one motion, sold the public a thing the public already had. Che cazzo?

9. Facebook (2012) — $16 billion. A debut so botched the Nasdaq’s systems failed at the open, and the stock spent a painful stretch below its offer price. Then the company got hammered for not having a mobile strategy, which spawned the doomscrolling era in which we now live. This was a classic version of what is now Big Tech’s standard share structure: dual classes of stock that locked control with the founders no matter how much equity the public bought — a governance design Musk has now taken to its limit.

10. General Motors (2010) — $15.8 billion. “Government Motors.” The federal government rescued GM with tens of billions in public money, steered it through bankruptcy, and then sold the public shares in the company it had just paid to save. The taxpayer funded the rescue and was offered the chance to buy in afterward, at market price.

Statecraft, legal fictions, debt offloads, bubble valuations, public bailouts turned back into IPOs — and one viable business. The list SpaceX just topped isn’t really a ranking of size. It’s a record of who got sold what pitch, and how long it lasted.

The machine underneath all of it was built exactly once, on a canal in Amsterdam, four hundred years ago. Here’s what it did the first time it ran.

Further Reading

• From Saudi Aramco to Alibaba: World’s biggest IPOs — the Reuters factbox ranking every offering by deal size

• How SpaceX’s Dream of a Record-Breaking IPO Stacks Up — Bloomberg on SpaceX against the top 100 listings since 2000

• SpaceX sets the stage for a record $75 billion IPO — CNN Business on the offering

• Biggest IPOs of All Time — Dealroom’s ranked list with deal sizes and dates

With a share price of $135 and a target of $75 billion at the start of trading Friday, SpaceX is now the largest public stock offering in the history of capitalism, and not by a little. SpaceX blows away the prior record-holder, Saudi Aramco, which raised $25.6 billion in 2019 and held the record for seven years. Alibaba, SoftBank, Visa, General Motors — the rest of the all-time list now sits a long way down the page.

The SpaceX numbers are mind-boggling. No wonder the Nasdaq bent its rules on stability and profitability and allowed an unproven behemoth like SPCX to join its ranks. And that, in turn, means that institutional funds — pension funds, index funds, your 401(k) — will automatically buy shares in the company if they’re tracking the Nasdaq-100.

But the numbers are especially disorienting when you consider what a share actually buys.

For one thing, it doesn’t buy any voice in what the company does. The New York Times profiled Justin Fishner-Wolfson, whose firm has spent fifteen years buying SpaceX shares and now holds more than 1 percent of it — about $20 billion at Friday’s price. He is among the most devoted investors in the company, having only ever bought, never sold. He’ll presumably be a multibillionaire as of today. But the Times also asked him whether he had ever given Musk tough feedback. His answer, which the reporter says followed a 19-second pause:

“You can say whatever you want to say, and then he’ll either change his mind or he won’t. It’s usually a short conversation.”

Twenty billion dollars of ownership buys nothing more than that. This is the essence of investing in Elon Musk, and in tech in general, and as we’ll see shortly, in the very first stock ever offered.

The terms are in the filing. Musk keeps more than 82 percent of the voting control after the offering, so every question is settled before a shareholder raises a hand. The $75 billion goes into the company and stays there; the only way to get your money back is to sell your share to someone else. And the company is losing billions of dollars a year, which means the value of what you bought rests almost entirely on what people believe it will be worth later.

Shares anyone can trade. Money the company keeps for good. Control held by one man. All of it wrapped around a business that several governments — including that of the United States — now depend on. This sort of arrangement has been sold to the public before. In fact, these were the conditions of the first stock offering in history. It ran for nearly two hundred years, and it ended badly.

Read more

Eleven years later, Hollier doesn’t have to answer that q…

Read more

On September 28, 2008, on Omelek Island in the Marshall Islands, the engine of a Falcon 1 rocket sat under the nervous gaze of hundreds of SpaceX engineers and, by extension, thousands of its investors. The first three attempts to ignite and launch the thing had failed — once when a fuel leak caused a fire at launch, once when a control system failed, once when an unexpected thrust surge backed the rocket violently into its own second stage. Rocket launches are the very definition of difficult, not to mention costly: the company had only enough money left for this attempt. Employees watched a live feed from the other side of the world, some of them crying during the countdown. And then…it lit. The Falcon 1 lifted off. And it burned cleanly for nine and a half minutes. For the first time in history, a private company’s liquid-fueled rocket left the grip of gravity and reached orbit.

The rockets did exactly what the company promised, delivering NASA’s payloads to orbit cheaply and reliably. (And even, to my utter amazement, returning to Earth upright for reuse.) But the real money came from another payload they carried. By the end of 2025, Starlink (SpaceX’s satellite internet constellation, delivered into orbit in stages until they ringed Earth) had grown to more than 9 million subscribers globally, generating the majority of SpaceX’s estimated $15.6 billion in annual revenue on a recurring, subscription-based model. The kind of economics that turn an impossibly expensive rocket company into a cash machine. A machine that in 2024 quietly turned a $791 million profit.

Then Musk looked at that machine and decided to use it to fund a war for AI dominance, one he was clearly losing.

In February 2026, SpaceX acquired xAI — Musk’s artificial intelligence company, maker of the Grok chatbot — in an all-stock deal. The acquisition brought xAI’s losses onto SpaceX’s consolidated books. Building, maintaining, and launching rockets is expensive. But in 2025 alone, xAI spent $12.7 billion in capital expenditures — more than the combined $8 billion SpaceX spent on its entire Starlink and rocket launch business. SpaceX posted a net loss of $4.94 billion for the year, a swing of more than $5 billion from the year before. The profitable company became the wallet for the unprofitable one. That’s what this week’s SpaceX IPO is asking the market to fund.

Before we get to whether this is a good bet, we need to be clear about what Musk is actually asking of you, his investors.

He wants your money. He does not want your opinion.

SpaceX is listing under a dual-class share structure. “Dual-class” means two tiers of stock. One class for outside investors, with standard voting rights. One class for Musk and insiders, with dramatically amplified voting power. According to SpaceX’s own S-1 filing, as reported by Reuters, Musk’s Class B super-voting shares give him 85.1% of the voting power of the entire company. You can buy a piece of SpaceX on Friday. You cannot tell SpaceX what to do. Ever.

Musk didn't invent this arrangement. When the New York Times and the Washington Post went public in 1967 and 1971 respectively, the Sulzberger and Graham families kept voting control through dual-class share structures, explicitly to protect editorial independence from market pressure. The argument was: trust us with your money, we'll protect the journalistic mission. But that was a narrow carve-out in a market that had largely rejected the practice since a Dodge Brothers stock scandal in 1926. What tech companies did was rehabilitate dual-class shares as a general business model, and the markets rewarded them for it. Google was the first tech giant to adopt the structure when it went public in 2004. Then Meta used one. Snap went all the way and offered public shareholders zero voting rights whatsoever. Investors piled aboard anyway. The stock prices went up. The model spread.

What sets SpaceX apart is its scale…and geopolitical impact. Google, Meta, and Snap trade in your data and attention. SpaceX’s customers include NASA, the U.S. Department of Defense, commercial satellite operators, and an increasing number of foreign governments. When Russia invaded Ukraine in 2022, Starlink became the communications backbone of a nation at war in ways no government had planned for and no competitor could replace. Musk held — and on at least one documented occasion, withheld — connectivity that determined the course of military operations. No shareholder vote required. No board approval sought.

Institutional Shareholder Services once described Facebook’s governance structure as having “a defense against everything except hubris.” Fidelity’s then-general counsel said companies adopting dual-class structures are “less likely to have alignment and less likely to have the accountability.” Those lines were written about companies that sell ads alongside your information diet. SpaceX sells rockets to the US government and internet access to armies.

Now, let’s be clear: The pre-2004 era of American corporate governance was not a golden age of socially responsible capitalism.

In 1998, the Rockefeller family led a shareholder revolt against ExxonMobil over its climate change policies. Sure, the Rockefellers were heirs to a fortune built on oil, but still, it was investor activism, and from truly powerful investors.

They lost.

In 2019, ExxonMobil shareholders rejected a proposal to create a special board committee on climate change. Voting rights, held by investors who had them, did not reliably produce good outcomes for the planet, for workers, or for the communities that bore the costs of corporate decisions.

But what pre-2004 governance did punish was bad management. In 1992, GM’s outside board members removed CEO Robert Stempel — the first GM chief to lose his job since the company’s founder was ousted in 1921 — after the company lost more than $4 billion in a single year. In 2004, a 43% no-confidence vote by Disney shareholders, coordinated by Institutional Shareholder Services, Glass Lewis, and CalPERS, stripped Michael Eisner of his chairmanship and ultimately ended his tenure. Both outcomes were possible because there was a structural mechanism that made them possible. A board with real power. Investors with real votes. Accountability that a CEO could not insulate himself from by design.

The dual-class era removed that form of governance, and called it a feature. I am not mourning the loss of a system that served us well. I am noting the removal of the last responsive brake pedal — just as a company that touches nearly every corner of our lives is asking the public markets to invest the kind of capital that, in any prior era, would have come with at least some form of structural check.

Let’s make the case for Musk’s offering, because it’s real and it deserves a fair hearing.

The guy has pulled it off before. Starlink looked financially deranged until it generated $15 billion in recurring revenue. Musk has a pattern, maybe even a strategy — build infrastructure at catastrophic cost, then wait for the world to need it. And xAI’s financials, as alarming as they look, contain at least one data point you should consider: in May 2026, Anthropic — one of xAI’s chief competitors in the frontier AI race — agreed to pay xAI $1.25 billion per month to access compute through its Colossus data center. A competitor paying nine figures every 30 days for your infrastructure is an important sign you’ve created something the world wants. Maybe even something the world needs.

If the bull case is right, xAI becomes what Starlink became: infrastructure so embedded in critical systems that governments and industries cannot function without it. The compute backbone for the AI industry. A space-based cloud that governments route their most sensitive communications through.

(I’m trying to put aside my broader democratic concerns, but let me describe them for a moment: A single private actor — unelected, unaccountable to shareholders, answerable to no board with meaningful authority — will have built and will control the infrastructure that militaries, governments, rival AI companies, and billions of ordinary users depend on. Ukraine showed us what it looks like when a nation discovers mid-war that it has handed that kind of dependency to one man. So I do not find the bull case for SpaceX reassuring. If anything, I find it more alarming. Anyway, back to the investors.)

Here’s what the S-1 says about Musk’s incentives. In January 2026, SpaceX’s board approved a compensation package that will award Musk 200 million super-voting restricted shares if the company hits a market value of $7.5 trillion and establishes a permanent human colony on Mars with at least 1 million people. A separate tranche awards 60.4 million additional shares for operating space-based data centers providing at least 100 terawatts of capacity. Executive compensation expert Eric Hoffmann of Farient Advisors told Reuters he knew of nothing remotely comparable at any other company. (He also noted something that should give every SpaceX investor pause: SpaceX and Tesla — both controlled by Musk — are now effectively competing against each other for his attention.)

As an investor, you’re not just betting that the infrastructure play works. You are betting that a man whose personal compensation is tied to colonizing Mars will somehow prioritize your returns along the way.

Two institutions have rendered quiet verdicts on this offering in recent weeks.

Denmark’s AkademikerPension has blacklisted SpaceX entirely, citing what it called the company’s “catastrophic governance structure.” The S&P 500 index committee — which had considered whether to waive its profitability, seasoning, and float requirements for very huge (“megacap”) companies ahead of this listing — declined to do so last week, meaning the world’s largest passive funds will not be forced to buy this stock.

Neither institution is concerned that the rockets don’t work. Neither worries the infrastructure bet can’t pay off. What they said, in the institutional language available to them, is that this company, at this valuation, run by one man accountable to no one, is not yet a company the world's most careful money should touch.

A few weeks ago, I sat in a federal courtroom in Oakland and watched Elon Musk testify.

He wasn’t there to defend a business decision. He was there because he had sued the people who, in his view, had stolen the founding narrative of the AI industry from him — the argument that he, not Sam Altman, not the other OpenAI founders, was supposed to be the person who built the future. The trial exposed, in real time, a man for whom the question of historical credit had become consuming. Not in the way that ambitious people are ambitious. (And these people, who literally consider their companies to be the most important thing in the world, redefine ambition.) In the way that people are when they feel something has been taken from them that they cannot get back.

xAI posted an operating loss of $2.47 billion on $818 million in revenue in the first quarter of 2026 alone. That could be the financial profile of a company following a very scary but very audacious plan. That could also be the financial profile of someone utterly and irrationally determined to not be second.

The investors who show up on Friday aren’t buying a share of SpaceX. They’re buying a bet on whether the man whose people wept watching that fourth rocket light is still the one making decisions.

And once his company goes public, whoever Musk is now will have more resources — and less accountability — than ever before.

Further Reading

• SpaceX S-1 filing, SEC EDGAR — the primary source document for the IPO’s governance structure, compensation terms, and voting rights disclosures

• Congressional Research Service: Dual-Class Stock: Background and Policy Debate — the nonpartisan congressional primer on dual-class share history, policy debate, and the long record of companies — from Ford to Coca-Cola to media empires — that have used the structure to insulate founders from accountability

• Fortune: Without WaPo, Graham Holdings Should Cede Control to Shareholders — on how the Washington Post, the New York Times, and Dow Jones all went public with dual-class share structures to protect family editorial control, and why critics said the rationale collapsed once the journalism mission was sold off; the Graham and Sulzberger families as the template SpaceX is now extending to rockets and satellites

• Fortune/CNN Money: Robert Stempel — The Star-Crossed Career of a Fallen GM CEO — on the 1992 boardroom coup that removed General Motors’ CEO after the company lost more than $4 billion in a single year; the first GM chief to lose his job since the company’s founder was ousted in 1921, and the clearest pre-dual-class example of investor accountability functioning as designed

• ABC News: Disney’s Eisner Rebuked in Shareholder Vote — contemporaneous March 3, 2004 coverage of the annual meeting at which 43% of Disney shareholders withheld votes from CEO Michael Eisner, forcing the board to strip him of the chairmanship the same day; the no-confidence vote was later revised upward to 45.3% by AP

• New York Review of Books: The Rockefeller Family Fund vs. ExxonMobil — written by the Rockefeller family members who led the decades-long shareholder campaign against ExxonMobil’s climate denial; documents why full voting rights still failed to force social accountability, making the case that the old governance model’s brake worked for financial misconduct but not environmental harm

• Reuters: SpaceX could value company at record-setting $1.75 trillion — on the IPO valuation and Musk’s 85.1% voting control, sourced directly to the S-1 filing

• Reuters: SpaceX ties Musk compensation to Mars colonization goal — on the compensation package that awards Musk super-voting shares contingent on reaching a $7.5 trillion valuation and establishing a Mars colony

• Reuters: Musk ordered Starlink shutdown during Ukraine’s 2022 Kherson counteroffensive — on how Musk personally ordered coverage cut during a Ukrainian military operation, disabling drones and disrupting artillery; SpaceX disputed the reporting

• Harvard Law School Forum: Dual-Class Shares: Governance Risks and Company Performance — on the academic research showing firm value tends to decline over time under dual-class structures, and the alignment gap between founder voting power and financial risk

• Morningstar: “Financials Look Reckless” — Lifting xAI’s Hood in the SpaceX IPO — on xAI’s burn rate and what the combined entity’s financials actually show

As you may know I became a contributor at CNN as of a few weeks ago, and they were kind enough to issue a press release about it yesterday, for which I’m proud and grateful — not in the horrible LinkedIn way, but the true emotions. Thank you especially to all of you for supporting the work that CNN has now decided deserves time on their air.

When he stood onstage with the Pope last month, Christopher Olah, cofounder of Anthropic and the only representative of a tech company there, listened to Leo XIV detail a framework of moral accountability for harvesting data, compressing choice, and automating death. As he nodded along with the pontiff, he must have known a few contradictory things he wasn't going to discuss that day. For one thing, his company's technology is embedded in the battlefield-decision systems the Pentagon has been using to choose targets in Iran. (And attacks on Iran were taking place that very afternoon.) And his company was about to file confidentially to go public, in what could be the largest IPO in the history of capitalism.

In his remarks, Olah told the assembled cardinals and theologians that every frontier AI lab — including his own — “operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing.” He said the world of technology needs “moral voices that the incentives cannot bend.”

Now, however, his company and its closest rival are about to experience the most powerful incentives capitalism has to offer, and their moral voices will be brutally tested — pushed, pulled, and twisted — by the market.

Anthropic filed its S-1 confidentially on Monday, according to The New York Times and others. The numbers are extraordinary: the company’s annualized revenue reached $47 billion in May 2026, up from roughly $10 billion the prior year — what CEO Dario Amodei has called a “crazy” trajectory of “80x growth.” The anticipated IPO valuation is approaching $965 billion, with Goldman Sachs, JPMorgan, and Morgan Stanley already in the room. The stock could debut as soon as the fall. Meanwhile, OpenAI, the company’s closest rival, is preparing its own listing, targeting above $1 trillion, as early as September. Each possibility is an unprecedented moneymaking opportunity for its founders, and for Wall Street.

Neither company was built for this. OpenAI was founded as a nonprofit research lab. Anthropic was created by former OpenAI researchers who left specifically because they thought the company was moving too fast without adequate safety commitments. Dario Amodei and his colleagues incorporated Anthropic as a Public Benefit Corporation — a legal structure explicitly designed to put mission alongside profit. They built a Long-Term Benefit Trust, an independent governance body with escalating board-election rights, meant to prevent short-term financial pressure from overriding safety decisions.

The Wall Street Journal has reported on what happens when that architecture meets public markets: “Going public would likely force Anthropic to choose between its current public benefit structure and shareholder obligations.” A Harvard Law review of the structure, published this week in Fortune, found that a supermajority of Anthropic’s investors can terminate the Trust and remove the directors it appointed. The safety architecture is real, but it turns out it can also be dismantled by its own people, given the right pressures.

Which brings us to military contracts.

Anthropic wasn’t strong-armed into working for the military. It chose to do so.

In November 2024, Anthropic partnered with Palantir and Amazon Web Services to supply Claude to US defense and intelligence systems, which deployed it inside classified environments. The partnership allowed Claude to operate within Palantir's AI Platform on Impact Level 6-accredited infrastructure — the highest security classification available to commercial vendors. Eight months later, Anthropic was awarded a $200 million DOD contract and became the first AI lab to integrate its models into mission workflows on classified networks. None of this was done with any outward sign of reluctance.

But the company did eventually find itself articulating some regret. After Anthropic learned that its technology had been used in planning the kidnapping of Nicolas Maduro from Venezuela, it told the DoD it couldn’t permit the use its tech for the mass surveillance of American citizens (foreign citizens were still fair game, in their view), nor as the sole decision-making method in choosing people to kill.

That’s where a standoff began. In February 2026, the Pentagon gave Anthropic a deadline: drop those restrictions, or lose the contract. Anthropic refused. Trump ordered the federal government to cease using its products. Defense Secretary Pete Hegseth designated the company a supply-chain risk — a label previously reserved for giants like Huawei, controlled by geopolitical rivals. Anthropic sued.

But the handy narrative — principled anti-war company vs. unscrupulous military — doesn’t reflect what actually took place. CEO Dario Amodei wasn’t anti-war, for one thing. In his public statement the day the Pentagon standoff reached its peak, Amodei wrote: “I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries.” He said Anthropic had been “the first frontier AI company to deploy our models in the US government’s classified networks, the first to deploy them at the National Laboratories, and the first to provide custom models for national security customers.” He listed all of this as a point of pride.

Anthropic also didn’t sue on ethical grounds, or to cut ties with the military. It sued to continue to be able to sell to the DoD. And the military wasn’t exactly able or willing to kick its Claude habit, either.

Hours after the ban was declared, the Pentagon was nonetheless using Claude in its first airstrikes on Iran. The Washington Post reported that to strike a thousand targets in the first 24 hours of Operation Epic Fury, the US military leveraged Claude through its integration with Palantir’s Maven Smart System — for intelligence assessments, target identification, and simulating battle scenarios. (Listen to my conversation here with Katrina Manson, the reporter who went deepest inside the story of Maven Smart System.) Pentagon Chief Information Officer Kirsten Davies confirmed to Sen. Jack Reed that “the system is active right now.”

The ethical lines Anthropic drew were precise, and, read quickly, highly principled: no mass surveillance of Americans, no fully autonomous weapons. But using Claude to help select which Iranian facilities to bomb was comfortably in-bounds.

Why walk into this weird, shady market? After all Anthropic’s first military contract was worth only $200 million. Small money, for a company on its way to a trillion-dollar valuation. But military contracts are relationships. They are dependencies. And they are a long-term source of potentially enormous amounts of reliable revenue for a company that is about to desperately need it.

Read more

According to an NBC News poll conducted this past March, AI is less popular with American voters than ICE. More specifically: only 26% of Americans hold favorable views of artificial intelligence, against 46% unfavorable — a net negative of twenty points. The agency that rolls up in the night to pull families apart rates better than the technology writing their children’s homework.

That’s the political backdrop for Sam Altman’s trip to Washington, DC. this week. He arrived in the capital Wednesday to meet with White House officials, and with House Speaker Mike Johnson and House Minority Leader Hakeem Jeffries — the broadest possible spread of political power, a deliberately bipartisan tour. He came with a message about safety, national security, and America’s need to lead on AI. And he came, in effect, to speak to the president, Congress, and — listening through the walls for any sound that might move a stock — the investors who will soon fund OpenAI’s IPO. Those three audiences do not want the same things from him. In some ways, they want directly opposing things. What he says to satisfy one of them becomes a liability for the other two. Here’s who he has to win over this week.

Trump

Eleven days before Altman’s Washington visit, OpenAI filed a confidential IPO registration statement with the SEC — Goldman Sachs and Morgan Stanley leading the process, a September listing targeted, a valuation above $1 trillion sought. The filing, first reported by the Wall Street Journal and confirmed by CNBC and the New York Times, landed just as a jury in Oakland dismissed Elon Musk’s lawsuit against the company, clearing the last major legal obstacle to going public. Now, two weeks later, Altman has come to Washington to endorse an executive order his company has spent weeks helping to shape, and to curry favor with the White House.

The order, signed by President Trump on Tuesday, establishes a voluntary framework under which AI companies would submit their most advanced models for up to 30 days of government review before public release. Altman posted six words on X: “The new EO gets the balance right.” This is the public face of a relationship Altman has been carefully constructing since before Trump’s inauguration — he attended the ceremony, donated $1 million to the inaugural fund, and committed OpenAI to the $500 billion Stargate AI infrastructure project, which broke ground on its first major campus in Saline Township, Michigan on June 1st, with Altman and Oracle’s Larry Ellison on stage together.

OpenAI seems to have had a hand in softening this EO. An earlier draft called for a 90-day review period for frontier models. The final version requires 30. Newsweek reported that “tech companies were said to have convinced the president to hold off on his previous plan,” though no outlet has specifically named OpenAI as the force that drove the reduction. What is confirmed, per Yahoo News, is that Altman and other OpenAI staff “spent several weeks in dialogue with officials as the policy took shape” and that OpenAI was among the companies working directly with the White House on the order’s language. In the end, Altman got a watered-down version of an already pretty thin piece of regulation: only 30 days’ notice, a voluntary compliance framework, and no third-party or federal testing requirements. Altman is presumably at the White House to shake hands and clap for all of that.

What he may also be there to push along is the deterioration of Trump’s relationship with Anthropic, OpenAI’s closest rival and once the favored provider of AI technology to the Pentagon. In late February, after Anthropic tried to insert into a Pentagon contract explicit prohibitions on mass surveillance and autonomous weapons, Trump directed federal agencies to stop using its tools and designated the company a supply-chain risk — a label typically reserved for foreign adversaries. Anthropic’s $200 million DoD contract evaporated in an afternoon. OpenAI signed its own, new Pentagon deal hours later. As we’ll discuss in a moment, OpenAI needs that military contract, and many more besides, and the White House is how he can best try to keep that faucet open. Because whatever Altman’s relationship with this administration looks like from the outside, he is still subject to the caprice and favoritism of the president, and he’s still just a vendor whose contract can be pulled.

Congress

Altman once had senators more or less eating out of his hand. In the spring of 2023, the night before his first Senate testimony on AI, Sam Altman arranged a large private dinner and ChatGPT demonstration for members of Congress. Even representative Ted Lieu of California, a Democrat who holds a computer science degree from Stanford and sits on the House AI caucus, came out describing the experience as genuinely shocking — something close to magic. The guests were clearly awed. The impression OpenAI’s CEO left on them was brilliantly constructed and highly effective: we need to understand this thing before we can govern it, and this wonder kid Altman explains it best.

That version of Washington does not exist anymore.

Read more

I’ve been in a strangely upbeat mood the last week. It’s undoubtedly a function of an early summer here in California, relaxed mornings as my kids sleep in rather than struggling to get out the door to school, a little bit of financial breathing room thanks to my new contributor role at CNN, and the arrival of stone fruit.

But I’m pinning it to the Pope.

As we covered here last week, Pope Leo XIV’s Magnifica Humanitas — his new teaching exhorting the world to fight off the worst temptations of A.I. — is a piece of truly thoughtful scholarship, attuned to some very subtle threats posed by the consolidation of power and compression of agency in this technological and political moment. And as I wrote in my final column for Hard Reset last week, it’s part of a modern tradition of Papal language making its way into real change for normal people.

Hard Reset

You Have Inherent Value: An Ancient Lesson About New Machines

Author’s Note: This is my final column on AI and labor for Hard Reset. (I’m grateful to them for giving me this space, and I have no doubt we’ll work together again soon.) It’s been a tremendous pleasure to cover the beat for you, as this has been a particularly thoughtful and active audience, full of people clearly working hard to sort out a way forwar…

Read more

a month ago · 13 likes · Jacob Ward

In that column I mentioned Monsignor John A. Ryan, the priest-economist who took the 1891 encyclical of the last Pope Leo — in which Leo XIII argued that human beings have God-given value — and turned it into an argument that they also have a God-given right to a decent wage. Through Ryan, those ideas eventually became law and helped inspire the New Deal.

So I’ve been geeked on language, and its power to make change. When our Pope Leo writes about the harvesting of “vital data which, once aggregated and analyzed, can be used to train predictive models, guide investment strategies, anticipate crises and, above all, determine who and what is deemed to matter,” he’s helping all of us name a real problem, and maybe — just maybe — start thinking about how to fight it.

The New Vocabulary We Need

I have several times pointed here to the book Exit, Voice, and Loyalty, by the economist Albert O. Hirschman, in which he argued that any functioning society/company/government has to give its people/customers/citizens the right to walk away, speak up, or stick it out. Without those three choices, we fall into something dark and dysfunctional. I love the concision of that argument — it’s a wonderfully short book — and the way it helps me articulate today why clicking “Yes” or “No” on a Terms of Service agreement before using a new iPhone just doesn’t quite feel like we’ve been given all the choices we deserve before letting that technology and that company into our lives.

I was listening to the Optimist Economy podcast this weekend (two wonderfully clear-eyed thinkers blow up assumptions about fairness and progress and richness each week, it’s great stuff), and they mentioned the German word Zugzwang, which describes the moment in a chess match when the only moves left to a player are ones that put her into an increasingly bad position.

The news has been full of college seniors booing their AI-apologist commencement speakers — you go, kids — because they’re clearly seeing only Zugzwang ahead of them. Zugzwang is exactly how it feels to work in a company where the people who refuse to use A.I. heavily will clearly be axed, but where using A.I. will clearly hasten the axing of more or less everyone eventually. Zugzwang is an improvement on the English term "hegemony" — the way the dominated get enlisted to prop up their own oppression — because even in the hell of hegemony you get to make a living.

I’d never heard the term before, so it got me looking into cultures that have experienced real oppression and real multigenerational struggle to get out of it, because, I reasoned, there have to be more great, proven terms for the stuff we’re facing right now. Here are a few choice terms I’ve found, from cultures familiar with this sort of trouble. The first few help describe the problem. The final ones describe a solution.

Read more

I woke up at 5am to read it, and it’s remarkable.

It names algorithmic systems making decisions about credit, jobs, and welfare benefits. It names data extraction from poor countries as a new form of colonialism. …

Read more

At 11:30am today in Rome, the Vatican published Magnifica Humanitas, its official moral teaching about protecting the value of human beings in the age of AI. For the Vatican, it’s part of a long tradition of pushing back against the efforts of the market to evaluate humans on nothing more than what we can produce. For critics, activists, and policymakers, it creates language for resisting the market of human output that AI is enshrining. For Anthropic, which had a representative present at the unveiling in Rome, it’s a tremendous marketing opportunity.

I woke up at 5am PT to get as much of it into my head as I could. Here’s what I’ve pulled from it so far.

The document’s official title translates roughly as “The Grandeur of Humanity.” It’s Leo XIV’s first encyclical, published on the 135th anniversary of the encyclical that first pulled the church into “social doctrine,” Rerum Novarum. In 1891, that encyclical’s author, Leo XIII, looked at the industrial revolution’s wreckage and declared that workers had dignity far beyond their economic utility. The Catholic labor organizations that followed, particularly in Belgium, Germany, and the United States, took that declaration as a warrant. It took another forty years, a depression, and a president willing to act — but the New Deal’s intellectual scaffolding included Catholic social teaching that began with what Leo XIII put on paper.

Leo XIV is trying to do the same thing. He has clearly thought about this deeply, and the Magnifica Humanitas is technically literate and structurally aware — I was genuinely impressed by the sophistication of his analysis of the problem of AI. The question is whether there’s a labor movement, a litigation strategy, a legislative coalition in what he’s published today. Let’s look through it together.

Read more

These are Pope Leo’s words, describing a moment when technology has reorganized power faster than any institution can respond, a world in which an enormous accumulation of private capital has outrun every legal, moral, and political framework meant to constrain it, a system in which the people doing the work have no protection and no recourse.

But these are not from the encyclical on A.I. being released tomorrow morning. And it’s not by today’s Pope, Leo XIV. It was signed on May 15, 1891, by Pope Leo XIII, when he decided it was the church’s moral responsibility to fight back against the brutal effects of industrial capitalism.

The threats that Rerum Novarum (“Of New Things”) warned against in 1891 were not abstract worries about the future. They were well-documented horrors that had already afflicted generations of people.

In 1832, a British parliamentarian named Michael Sadler convened a committee to take testimony from factory workers, and what they heard constituted one of the most damning records of institutional failure in democratic history.

Matthew Crabtree was twenty-two years old when he testified. He told his interviewers that he’d started factory work at age eight. His shift was at least six in the morning to eight at night, with one hour off at noon. In busy periods, his bosses extended the shift an extra hour on each end of the day. He walked two miles to and from the job. Eight years old. When the committee asked how he woke up in time, he answered: “I seldom did awake spontaneously; I was most generally awoke or lifted out of bed, sometimes asleep, by my parents.” When they asked what happened when he was late: “I was most commonly beaten. Very severely, I thought.”

Hundreds of similar accounts emerged, and Crabtree’s testimony turned out to be typical. The report’s own editor noted the testimony was “picked almost at random from a bulky volume,” and a mid-twentieth-century historian described it as “a mass of evidence, constituting a most formidable indictment of factory conditions.”

Parliament read it and declined to legislate.

In fairness, it was easy for the wealthy to be blind to what was happening to working people. The urban planning of cities in England at that time hid the privations of the poor. Friedrich Engels, living in Manchester in the 1840s while working in his family’s textile firm, documented how the city had “contrived in its curious lay-out and hypocritical town planning to shield its appalling poverty from the prosperous factory owners and merchants” — the wealthy traveled from their homes to their offices without ever seeing the conditions on either side of the road. (I think here of the pristine roads of the South Bay cities that constitute Silicon Valley, and of the enormous encampment of homeless Californians huddled beneath the final approach to San Jose Airport.)

But the data alone should have been enough to spur regulatory action. In the central parishes of cities like Manchester, Liverpool, and Glasgow during the height of industrialization, average life expectancy had dropped to roughly twenty-five years. The Industrial Revolution had clearly redefined the value of humans, one based entirely on what they could produce in a waking day, even if it changed the number of years they were on this earth to be part of this terrible new output market. It was this redefining of the value of human beings that Pope Leo XIII wrote his encyclical to fight back against.

Tomorrow, the Catholic Church publishes Magnifica Humanitas: On the Protection of Human Dignity in the Age of Artificial Intelligence. Pope Leo XIV signed it on May 15th — the exact 135th anniversary of Rerum Novarum, Pope Leo XIII’s 1891 encyclical on labor and capital, written at the height of the first Industrial Revolution. It turns out that our Pope is an AI critic and reformer — and that it's central to his papacy.

Two days after his election last May, the new pope addressed the College of Cardinals and explained why he had chosen the name Leo. “There are different reasons for this,” he said, “but mainly because Pope Leo XIII in his historic Encyclical Rerum Novarum addressed the social question in the context of the first great industrial revolution. In our own day, the Church offers to everyone the treasury of her social teaching in response to another industrial revolution and to developments in the field of artificial intelligence that pose new challenges for the defence of human dignity, justice and labour.” He has clearly been thinking about this for a while

.The Vatican press director confirmed the framing: the name was “a direct recall of the social doctrine of the church and of the pope that initiated the modern social doctrine of the church,” and a reference to “men and women and their work, also in the time of artificial intelligence.”

Robert Prevost — before he became Leo XIV — learned about structural threats to human dignity firsthand. He understands how systems work, sure, having earned a degree in mathematics at Villanova in 1977. But he was ordained in 1982 and went immediately to Peru, to the mission fields in the north, in Chulucanas — a semiarid region near Ecuador, in the foothills of the Andes, where campesino and Indigenous communities lived in deep poverty and deep faith. He was there, watching people struggle under the Shining Path insurgency, when Peruvian democracy fractured and stumbled. After a comparatively cushy decade-plus in Rome, he went back to Peru as bishop of Chiclayo from 2014 to 2023, overseeing a diocese facing structural poverty, natural disasters, a pandemic, and 1.5 million Venezuelan migrants arriving with nothing. During the floods of 2022 and 2023, as a man in his 60s, he drove a van himself, transporting supplies to rural villages. Can one say of a Pope that he’s badass?

Leo XIII developed his convictions via an easier route. In 1891, he came to Rerum Novarum through thirty years of theological correspondence and pastoral letters — a Vatican diplomat who encountered industrial capitalism from the outside and theorized about it in exile. Today, Leo XIV has arrived at Magnifica Humanitas through something closer to the Sadler Commission testimony: contact with crushed people seeking dignity, and a firsthand experience of how fragile democracy really is.

What can we expect to see in what he’ll reveal tomorrow? His most recent public statement on AI came at Sapienza University of Rome on May 14th, the day before he signed the encyclical. Addressing students, he condemned the use of AI in warfare — “What is happening in Ukraine, in Gaza and the Palestinian territories, in Lebanon and in Iran describes the inhuman evolution of the relationship between war and new technologies” — and urged vigilance over systems “so that they do not remove responsibility from human choices and do not worsen the tragic nature of conflicts.” To the students themselves, he offered a remarkable one-line description of the value of human beings: “We are a desire, not an algorithm.”

The Vatican has engaged AI directly before. In January 2025, the Dicastery for the Doctrine of the Faith published Antiqua et Nova, a 118-paragraph doctrinal note on the relationship between artificial and human intelligence. Its central philosophical claim was pointed: that AI’s “intelligence” is evaluated “on the basis of its capacity to produce appropriate responses... regardless of the way those responses are generated” — a very astute functional definition that, the document warned, risks creeping from how we assess machines into how we assess people. Drawing an overly close equivalence between AI and human intelligence, it argued, “risks succumbing to a functionalist perspective, where people are valued based on the work they can perform. However, a person’s worth does not depend on possessing specific skills, cognitive and technological achievements, or individual success, but on the person’s inherent dignity, grounded in being created in the image of God.”

Rerum Novarum made the same argument 135 years ago about the effect of factory looms on perceived human value. But what it actually accomplished is up for debate, and if the AI encyclical is going to have a real effect, it will have to do better.

The depressing institutional response to the Sadler Commission is important to think on as Magnifica Humanitas goes out tomorrow. Parliament read that terrible testimony: sleeping children woken for 14 hours at a factory, meager food they were often too tired to eat, parents with no financial choice but to feed their own kids into the market’s maw. And then the political class quickly acted to soften it. They commissioned a second inquiry — the Althorp Commission — with factory owners on the panel this time. That inquiry produced more industry-friendly findings, of course. The Factory Act of 1833 that followed was underfunded, largely unenforced, and widely evaded.

We are in a similar moment. In 2025, after making noises about the need for AI regulation, OpenAI CEO Sam Altman appeared before the Senate Commerce Committee for a hearing titled “Winning the AI Race: Strengthening U.S. Capabilities in Computing and Innovation.” He told that committee that prior regulatory approval for AI innovations would be “disastrous.” Last week, as Politico reports, President Trump scrapped a planned signing of an executive order implementing a federal pre-review process for AI models after AI leaders balked and his former AI czar David Sacks carried Silicon Valley’s concerns to Trump in a phone call. To this day, no comprehensive federal AI legislation has passed.

In 2026, state legislatures have published more than 2,000 AI bills — a patchwork filling the federal void, precisely as labor law was enforced unevenly across jurisdictions in the 1840s. The states with the most protective laws are already retreating: Colorado recently rolled back its audit mandates; Connecticut’s broad bill died in the House.

What would it take for this encyclical to break that pattern?

Read more

When Google filed to go public in April 2004, Larry Page and Sergey Brin opened their IPO prospectus with a declaration. “Google is not a conventional company,” they wrote. “We do not intend to become one.” They promised to “optimize for the long term rather than trying to produce smooth earnings for each quarter” and to prioritize doing good over maximizing profit. The letter became a touchstone — proof, people said, that a technology company could go public without surrendering its soul.

By the time Page returned as CEO in 2011 and addressed his first earnings call, the register had shifted. “We’re very careful stewards of shareholder money,” he told analysts.

The mechanism that produces this change is slow and subtle, like a descending escalator: quarterly earnings calls create pressure to explain variance; analyst coverage creates pressure to hit estimates; short sellers create pressure to defend valuation. The principles don’t disappear. They just gradually get tucked deeper inside the company, and cease to be load-bearing.

The AI industry is about to test its commitments in that very way, and OpenAI seems to be going first.

The Wall Street Journal reports that OpenAI is preparing to file for an initial public offering, working with Goldman Sachs and Morgan Stanley on a draft prospectus, targeting a public debut as early as September. (The New York Times confirmed the Journal’s account.) OpenAI is valued at $852 billion in its most recent funding round. A listing at that valuation would be the largest IPO in American history, and would place it next to the value of peak dot-com Microsoft or that of Standard Oil just before its 1911 antitrust breakup.

Why now? Well, two days ago Elon Musk’s lawsuit, which demanded that the company be converted back to a nonprofit, was tossed out. (The jury never ruled on whether OpenAI had abandoned its founding promise — to develop artificial intelligence for the benefit of humanity — nor on whether Sam Altman was a trustworthy steward of that promise. It ruled that he’d sued too late, and that was that.) Judging from the timing, and in spite of Musk’s plans to appeal, that presumably means the company considers the lawsuit to have been their last obstacle, and also considers it to be out of the way.

OpenAI and much of the business press seem to see this moment as vindication. The legal threat to its for-profit structure is gone. Public capital markets — and the cash they can deliver — are now within reach. Sam Altman, who built the company from a nonprofit research lab into one of the most valuable private companies on earth, will have liquid shares to show for it. By almost every conventional measure, this is what success looks like.

But what will success cost a company built on the promise that it would be different? And what will it go on to cost all of us?

Read more

Read more

Nope. In not even an hour and a half, they ruled that Elon Musk had filed his lawsuit too late. The judge threw out the rest of the case as a result. Time to go home, everyone. Thanks for your service.

I’ve been covering this trial from inside the Oakland courthouse since it began — standing in line at 5:30 a.m. for one of the thirty public seats, watching Musk testify about his efforts to take control of OpenAI even as he says he believed (and perhaps hoped) it would change the world, sitting there when Sam Altman was asked within the first two minutes of cross-examination, “Do you tell lies?” It wasn’t just about the billionaire feud, though. It was a rare chance for reporters to see inside the companies, and inside the communications of the tiny handful of people determining the future for all of us.

First, let’s get reacquainted with the case.

Musk sued Altman and OpenAI in February 2024. He put $38 million into OpenAI’s early years on the understanding, he said, that the company would remain a nonprofit dedicated to building AI safely for humanity. When OpenAI created a for-profit subsidiary in 2019, and then struck a $10 billion deal with Microsoft in 2023, Musk said the charity had been looted. He wanted $134 billion in “ill-gotten gains” returned to the OpenAI foundation, Altman and Brockman removed from leadership, and the for-profit restructuring unwound.

The case rested on documents that had never been public before — emails and texts and Greg Brockman’s personal diary entries, subpoenaed during discovery. As I reported, those documents showed the people at the top of the AI industry writing to each other with a casual assumption of world-historical importance. In one email, Musk mused about whether OpenAI would need a twelve-person board or a sixteen-person board — the difference being how much of the fate of humanity was riding on the company.

The testimony that followed was equally illuminating. Ilya Sutskever — the researcher most responsible for what generative AI can actually do — confirmed under oath that he spent a year building a 52-page dossier documenting Altman’s “consistent pattern of lying.” Former CTO Mira Murati, who determined how the technology feels to use, accused Altman of dishonesty. Former board members Helen Toner and Tasha McCauley said the same. Dario Amodei, who left OpenAI to found Anthropic, had said that Altman had misrepresented the terms of the Microsoft investment. As I wrote at the time, the people who worked most closely with Altman described the same pattern under oath. Altman himself was forced to admit that he has lied in his life, sure, although he denied doing it for business advantage.

The jury decided that none of that was the question in front of them. The question was: did Musk know about OpenAI’s shift toward profit-seeking in time to file a lawsuit, and did he wait too long? The jury found he did. Judge Gonzalez Rogers agreed, saying “there’s a substantial amount of evidence to support the jury’s finding.” (The jury’s verdict was to be only advisory, and Judge Gonzalez Rogers would have been the one making the final call under enormous scrutiny — its decision must have come as some relief to her.)

Musk’s team says they’ll appeal. His attorney Steven Molo told reporters the verdict was a narrow decision on “technical legal issues” and that Musk’s side had proved the core of its case. Musk, who skipped the end of the trial for his trip to China with President Trump, posted that Judge Gonzalez Rogers is a “terrible activist” judge and wrote that she “just handed out a free license to loot charities if you can keep the looting quiet for a few years.”

OpenAI’s attorney William Savitt — a very good lawyer who deserves his professional satisfaction today — told reporters outside the courthouse that this was “a hypocritical attempt to sabotage a competitor.”

Both of those framings are, in their way, accurate. And neither of them addresses the more durable problem.

Read more

Paid subscribers get early access to this and all my analysis, as well as written reports, including all the source documents.

So why get on the plane?

When China joined the World Trade Organization in December 2001, it transformed global business. The deal — backed by the Clinton administration, and celebrated (fists pumping, corks popping) in the boardrooms of American technology companies — promised mutual benefit through interconnection. Open markets, shared infrastructure, growth without friction. The companies that would one day be known as Big Tech were barely large enough to understand what it was about to do for them. But what they were building — platforms, supply chains, chip architectures — turned out to be perfect for the world the WTO agreement was creating.

That world lasted about twenty years.

On Wednesday, Air Force One touched down in a new world. It landed at Beijing Capital International Airport carrying President Trump, Secretary of State Marco Rubio, Defense Secretary Pete Hegseth, U.S. Trade Representative Jamieson Greer — and the CEOs of Apple, Nvidia, and Tesla.

Tim Cook, Elon Musk, and Jensen Huang joined Trump’s first visit to Beijing since 2017 alongside the CEOs of Blackrock, Citi, Meta, and others. (Huang joined as a last-minute addition in Alaska, after Trump called to get him literally on board — Pull over, we gotta grab Jensen.) Musk even brought his son X Æ A-XII to the Great Hall of the People.

Quick aside: Dina Powell McCormick, President of Meta (and one of only two women on the trip) is notable in that she’s there in place of Mark Zuckerberg, whose desperate desire to get his products into China once led him to offer Xi Jinping the chance to name Zuckerberg’s son. (Weird offer. Xi refused.) The American uber-capitalists’ pilgrimage to China is a well-worn one.

At a state banquet Wednesday evening, Xi Jinping told the assembled American executives that China’s door would “open wider,” according to state-backed media Xinhua, observing that U.S. companies are “deeply involved in China’s reform and opening up, and both sides have benefited.”

That’s a gracious way of putting it. It also fails to describe the pickle these three CEOs, and the rest of the American tech sector, are in.

The three men who flew to Beijing aren’t looking to expand their business in China. They’re trying to keep from losing it.

Read more

Read more

Subscribe now

A mishmash of lawyers, reporters, and a few protesters watched as Sam Altman was finally sworn in Tuesday morning in a federal courtroom in Oakland to defend himself against accusations of tricking the world’s richest man into funding a sneaky bid to create the world’s most powerful company. I was there.

The case, in its narrowest legal form, asks whether Altman and OpenAI breached a charitable trust by converting a nonprofit into a profit-seeking enterprise. Musk says they promised him something different. Altman says they didn’t.

But as I explained to a half-dozen CNN producers today texting me to see whether they should put this stuff on the air, the testimony wasn’t really about contract law, at least not for me. It was an anthropological exhibit, a power trip, a ringside seat to the billionaire circus. (The press conference on the courthouse steps was framed by a half-dozen elderly women singing “We’ll Stop A.I., Hallelujah,” which I loved, and I’ve discovered that my kink is watching billionaires say “yes, your honor” to Judge Yvette Gonzalez-Rogers.) Two of the most powerful men in technology, both convinced they’re saving civilization, are in court over a falling-out that’s been building for nearly a decade — and in the process, they’re giving us all a portrait of what it looks like when very, very rich men who claim to be steering the future have to account for themselves.

Here’s what stood out.

1. Molo opened the cross-examination with one question: “Are you completely trustworthy?”

Steven Molo is Musk’s attorney, and the dude is not subtle. After a mannered and pleasant direct examination of Altman by his own lawyer, William Savitt, Molo came right in with the big questions: is Altman trustworthy? (You can read about all the people he’s worked with who say otherwise here.) Within the first two minutes of cross-examination, right after exchanging hellos, he asked Altman this:

Molo: Are you completely trustworthy?

Altman: I believe so.

Molo: Do you always tell the truth?

Altman: I believe I’m a truthful person.

Molo: That wasn’t my question, sir. Do you always tell the truth?

Altman: I’m sure there is some time in my life when I have not.

Molo: Do you tell lies to advance your business interests?

Altman: No. No.

Molo: Have you misled people with whom you do business?

Altman: I believe I am an honest and trustworthy business person.

Molo: That wasn’t my question about what you believe. Have you misled people with whom you do business?

Altman: I do not think so.

Molo: Would they think so?

Altman: I can’t answer that for other people.

Gotta hand it to him: Altman was calm and pleasant throughout, while Molo became (or at least acted) angrier and angrier. It wasn’t clear to me whether Molo was honestly irritated, but he couldn’t seem to make Altman angry. Altman maintained the affect of a man who has been asked tougher questions than this. It made me wonder what it would take to accuse Altman of something he’d take offense to.

2. Then Molo read him the list.

What followed was a methodical recitation of every person who has, under oath or on record, called Sam Altman dishonest. It took a while.

Ilya Sutskever, OpenAI co-founder, testified that Altman exhibited “a pattern of lying.” Mira Murati, former CTO, accused him of dishonesty. Helen Toner, former board member, accused him of lying about OpenAI’s safety processes. Tasha McCauley, also a former board member, testified to “a toxic culture of lying.” Soo Yoon accused him of “a lack of concern for the consequences of misleading others.” Dario Amodei accused him of misrepresenting the Microsoft investment terms.

Altman’s answer to most of these was a variation on: I didn’t hear that testimony. Because, after all, as CEO of an $850 billion company, he has other things going on. He couldn’t be in court every day.

Molo: Is it important to you to find out what’s going on in this trial?

Altman: Yes. Although I also have a very busy day job and have not been able to be here every day.

On Dario Amodei specifically, he offered the sharpest line of the day, as close to admitting he doesn’t like someone as I could imagine him saying:

“Dario accuses me of many things.”

3. Altman was entertaining mind-blowingly casual political ambitions in 2017.

While negotiating who would control the technology that might one day produce artificial general intelligence, Altman’s own colleagues — Greg Brockman and Ilya Sutskever — sent him an email asking why the CEO title mattered so much to him. Their email asked, specifically, about his “political goals.”

Molo asked if Altman had told them he wanted to be president of the United States.

Altman: I was thinking about running for governor at the time, which I believe is what they meant here.

Chill, bro. Just governor.

4. The warmest memory he had of Elon Musk was a late-night meeting in 2018 that ended with watching Musk scroll his phone.

Savitt, on direct examination, was walking Altman through the 2018 period when relations with Musk were still, relatively speaking, functional. Altman described a meeting at Tesla headquarters — about the structure of what would become OpenAI’s for-profit entity, the future of AI governance, billions of dollars — and how it had gone unusually well.

“This was, like, a pretty late night meeting, and then it was a long conversation about him showing us memes on his phone. It was just like him showing us memes on his phone. So that was, like, the happiest and calm part.”

Ah, the good times.

5. Shivon Zilis remaining on the OpenAI board after he discovered she was the mother of Musk’s children was, in Altman’s words, “a close call.”

Read more