The episode opens with a pre-roll ad from OneTrust, which positions itself as the solution to the tension between AI speed and organizational risk. The ad emphasizes that competitive pressure and board expectations are pushing companies to adopt AI quickly, but without governance guardrails, that urgency becomes liability. Listeners are directed to onetrust.com/ai.

Before a single news story is discussed, Casey Newton delivers a running monologue about his terrible week with technology: a CPAP machine installed by a concerned partner, a laptop destroyed by a water bottle in his bag, and a debit card linked to his transit card that expired and couldn't be replaced through two days of city bureaucracy. The result was a 12-minute Uber ride in a car that 'appeared to have been smoking since 1976.' Kevin Roose responds with deadpan sympathy and agrees to carry the episode — setting up one of the show's warmest moments as Casey declares Kevin his 'own personal Jesus.' The exchange captures the chemistry at the heart of Hard Fork and provides a breezy on-ramp before an episode packed with heavy financial and scientific news.

After the comedic preamble, the show formally begins. Kevin Roose identifies himself as tech columnist for The New York Times and Casey Newton as founder of Platformer. They preview three segments: the coming wave of AI IPOs involving SpaceX, Anthropic, and OpenAI; an interview with Kevin Hartnett about AI's impact on mathematics; and the rapid-fire HatGPT news riff. The tone shifts from playful to substantive.

The hosts plunge into what could be the most consequential IPO wave in financial history. SpaceX is furthest along: it aims to raise $75 billion at $135 per share, valuing the combined entity — rockets, Starlink, xAI, and X — between $1.75 and $2 trillion ^{[1] — Casey Newton "SpaceX IPO at $135/share: SpaceX plans to sell shares at $135 each, aiming to raise $75 billion and value the company at $1.75–$2 trillion." 04:17 Play} . Casey frames SpaceX as two great businesses (reusable rockets, Starlink) yoked to two terrible ones (xAI and X), which Elon Musk apparently folded in to hide losses from weaker assets ^{[2] — Casey Newton "It seemed like Elon Musk decided he needed to kind of hide his losses somewhere. And so they inherited the two worst companies he owned." 06:56 Play} . Anthropic, meanwhile, filed a confidential S-1 and is expected to IPO at over $1 trillion — an almost incomprehensible trajectory given Kevin Roose's recollection of visiting the company in 2023 when it 'actively resisted the idea of making money.' The company's annualized revenue run rate reportedly jumped from $1 billion in early 2025 to $50 billion. OpenAI is also preparing to file. The hosts also flag their personal conflicts of interest: The New York Times is suing OpenAI, and Casey's fiancée works at Anthropic.

With the mechanics of the IPOs established, the hosts zoom out to examine what this wealth concentration means for the city at the epicenter. Casey Newton describes a palpable shift from the abundance mentality he encountered when he arrived in San Francisco in 2010 — when anyone could start a company and prosper — to a new scarcity mentality where employees at most tech firms are now measuring themselves against early Anthropic and OpenAI staff who stand to become centimillionaires ^{[1] — Casey Newton "When you move into a world where there is what essentially amounts to a handful of lottery winners, and those are the only people that trul…" 12:22 Play} . Kevin Roose agrees, noting that even well-compensated engineers who 'thought they had made it' are now feeling precarious. Casey crystallizes the anxiety: a society where only a handful of lottery winners get to live the life they want produces 'massive social instability.' Kevin adds a personal note: he called it correctly in a year-end prediction episode, forecasting that 2026 would be the last year to buy a house in San Francisco as the real estate market goes haywire, with some homes now being listed with asking prices denominated in Anthropic or OpenAI stock.

Kevin Roose introduces a dimension of the AI IPO story that most outsiders miss: the sheer volume of philanthropic capital about to be unlocked. He references Nan Ransohoff's essay framing these IPOs as the 'third wave of philanthropy,' where tens or hundreds of billions could flood into charitable causes. The specific mechanics at Anthropic are striking: all eight co-founders pledged to give at least 80% of their personal wealth to charity, representing hundreds of billions of dollars at post-IPO valuations ^{[1] — Kevin Roose "Anthropic co-founders pledged 80% to charity: All eight Anthropic co-founders pledged to give at least 80% of their wealth to charity, pote…" 16:01 Play} . On top of that, Anthropic's employee stock-matching program offered early employees up to 3-to-1 matches for charitable pledges ^{[2] — Kevin Roose "Anthropic employee stock matched 3-to-1: Anthropic matched early employees' charitable stock pledges at a 3-to-1 ratio, dramatically amplif…" 16:20 Play} . Kevin Roose concludes that the combined effect could produce more philanthropic capital per year than the Gates Foundation — flowing primarily toward AI safety, global health, pandemic prevention, and effective altruism cause areas that the outside world might find 'fairly weird,' including, in Kevin's joking formulation, 'a great year to be a shrimp.'

The final stretch of the IPO discussion turns to governance and safety. Kevin Roose makes the case for worry: both OpenAI and Anthropic were founded by people specifically concerned that for-profit corporations could not develop AI safely, and going public adds activist investors and stock market pressure on top of every existing incentive to race ^{[1] — Kevin Roose "AI safety was the founding concern at both OpenAI and Anthropic, but going public adds a new force: the public markets will now pressure th…" 18:05 Play} . The public benefit corporation designation provides some legal buffer, but Kevin argues it won't survive direct confrontation with public market dynamics when a truly dangerous model is on the table. Casey Newton offers two counter-arguments: first, shareholder pressure can cut both ways — releasing a bioweapon-enabling model exposes a company to securities fraud suits; second, IPOs introduce mandatory financial disclosure, earnings reports, and shareholder votes that represent more meaningful democratic oversight than any current regulatory framework ^{[2] — Casey Newton "It's not just good. It's necessary. We cannot have a very small handful of companies that are growing this quickly, that are concentrating …" 24:31 Play} . Casey closes with a normative argument: concentrating this much wealth and power in a handful of private companies is unsustainable — sharing the upside through public markets is not just good, it's necessary. Kevin endorses the idea of retail investors being able to buy AI exposure, especially in a world where Bitcoin ETFs and military-member prediction market betting are already legal.

In a brief coda to the IPO segment, Kevin Roose shares a personal anecdote: he flew on a Starlink-equipped United Airlines flight and experienced 200+ megabits of in-flight internet for the first time. His conversion from skeptic to believer was immediate. Casey explains the marketing genius of United's deal: Starlink charges the airline, not passengers, so every flier experiences Starlink as a 'free miracle' — a genius introduction to the product that makes the value proposition immediately visceral. Kevin declares that when people get a taste of what Starlink offers, they will never go back, and predicts Starlink could become the largest company in the world.

The episode pauses for a mid-roll ad break covering three sponsors. OneTrust repeats its AI-ready governance pitch. Framer promotes its AI-assisted web design tool with a 30% discount at framer.com/hardfork. KPMG describes its 'client zero' model — embedding AI across its own enterprise first to build a blueprint for helping other organizations scale AI capabilities confidently.

Kevin Hartnett, former senior reporter for Quanta Magazine and current editorial lead at Cursor, joins the show to contextualize the rapid development of AI math capabilities. He recounts the arc from ChatGPT's early failures at basic arithmetic — mathematicians passed around examples of it claiming there are finitely many primes — to the IMO gold medal scores achieved by Google DeepMind, OpenAI, and Harmonic in 2025 ^{[1] — Kevin Roose "AI math: IMO gold medal score: Three AI labs — Google DeepMind, OpenAI, and Harmonic — achieved gold medal scores at the International Math…" 31:38 Play} . But Hartnett provides crucial context: even gold-medal performance at the IMO is 'essentially high school math, the hardest high school math in the world, but still just high school math' — representing roughly 0% of the way to frontier research mathematics ^{[2] — Kevin Hartnett "AI at IMO = 0% of frontier math: Kevin Hartnett argued that even achieving a gold medal at the IMO represents essentially zero percent of t…" 33:16 Play} . He explains why AI labs focused on math: the belief that teaching a model to reason through mathematical problems develops general-purpose logical reasoning that unlocks commercially valuable capabilities. The hosts also walk through the Erdős problem landscape: Paul Erdős compiled over 1,200 open mathematical problems with small cash bounties, and AI labs have been systematically working through them as capability benchmarks — though Hartnett notes most Erdős problems were not considered important by the mathematical community, more like 'sophisticated Wordle puzzles.'

The discussion of Erdős benchmarks builds to its climax: OpenAI's resolution of the unit distance conjecture. Hartnett explains what made this result different from the string of solved Erdős problems that preceded it. Other solved problems had been largely ignored by serious mathematicians — 'sophisticated riddles,' in Hartnett's framing — but the unit distance conjecture was one that many mathematicians had actively tried to solve, failed, and moved on from ^{[1] — Kevin Hartnett "OpenAI's proof of the unit distance conjecture wasn't just another solved Erdős problem. Mathematicians had actually tried to crack it; the…" 38:20 Play} . The methods OpenAI's model used were sophisticated, non-obvious, and genuinely surprising. Most importantly, the mathematical community agreed nearly unanimously that the result was publishable in the Annals of Mathematics, the top journal in the field. Hartnett describes the pattern that had developed before this proof: 'AI did this, but it can't do that. Oh, it did that? But still can't do this.' The unit distance conjecture proof ended that goalpost-shifting. AI, Hartnett concludes, 'can do absolutely top-tier research.'

Hartnett offers a memorable vignette from the Institute for Advanced Study in Princeton — described as 'the citadel of modern math.' In one afternoon, he encountered two top-tier mathematicians with completely opposite views. The first had just closed Gemini in frustration after it told him something demonstrably false. The second, superficially similar on paper, predicted that AI would put mathematicians out of business within 2 years because it would soon be strictly better than humans at all of it. Terence Tao — the greatest living mathematician and most important figure in the debate — sits squarely in the middle ^{[1] — Kevin Hartnett "At the Institute for Advanced Study, Kevin Hartnett met two top mathematicians in one afternoon with polar opposite views — one dismissed A…" 40:18 Play} . Tao's view, which Kevin Roose describes from a recent OpenAI video, is that AI is like an 'Iron Man suit' or a 'jetpack for thoughts': it allows mathematicians to test many more ideas quickly, radically lowering cognitive friction for exploratory work. Hartnett suspects the dismissive camp was in the majority a year ago but is shrinking rapidly.

The conversation turns to the Leiden Declaration, the open letter that prompted this interview. Kevin Roose characterizes it as 'a very worried document,' and Hartnett agrees: it reflects the anxiety of a discipline that successfully self-regulated for centuries now facing a massive exogenous disruption it had no part in creating ^{[1] — Kevin Hartnett "Over 800 mathematicians signed the Leiden Declaration warning that AI is producing plausible but unreliable proofs, eroding human expertise…" 42:37 Play} . The declaration has two main aims: first, setting rules for disclosure — if you use AI in a proof, you must say so, and the Math Archive is already enforcing this with year-long bans for undisclosed AI use; second, protecting the field's priorities from being overrun by whatever kinds of problems AI happens to be good at. Casey Newton raises the obvious counterargument: isn't this what every profession says when a new technology arrives? Hartnett concedes the point but argues that math is different — it's genuinely possible for AI to strip the incentive and value out of the discipline without fully replacing it, the way streaming can devalue the album without replacing the experience of listening to one. The deeper worry, Hartnett argues, is that mathematics is 'a quintessentially human endeavor, the pinnacle of human thought,' and that losing the human wrestle behind a proof changes the nature of the activity even if the outputs look the same.

The episode's intellectual centerpiece closes with Hartnett reflecting on where the story ends. Kevin Roose floats the idea that mathematicians might embrace AI as a way to accelerate toward the 'teleological end' of their discipline — finally understanding all of mathematics. Hartnett gently deflates this: mathematicians are not worried about running out of math to discover, and AI would have to be orders of magnitude more capable to make a dent. On the bigger question of whether mathematics as a human discipline will survive, Hartnett refuses to predict complete replacement. Something that has been so central to human activity for so long, he argues, will not simply 'disappear and be replaced by pushing a button.' He expects math to look radically different and places himself in the Terry Tao middle camp: human beings directing machines, choosing which problems to pursue, will remain important. The field will adapt — as almost every field must — and something impressive will emerge from the other side.

A short mid-episode ad block features three sponsors: OneTrust repeats its AI governance pitch; Comcast Business highlights SD-WAN, SASE, and AI-powered networking at enterprise scale; and Framer promotes its AI-assisted web design platform with a 30% discount for Hard Fork listeners.

The show's rapid-fire news segment kicks off with two stories. First: The Bot Company allegedly rented a San Francisco Airbnb under false pretenses, installed robots, ran training sessions for 11 days, and left the property damaged ^{[1] — Kevin Roose "The Bot Company rented an Airbnb in San Francisco under false pretenses, brought in robots in black cases, trained them doing household cho…" 53:33 Play} . The homeowner is suing for $12,383.50 — which Casey Newton, invoking his 'ALAB' (All Landlords Are Bastards) theory, considers a perfectly normal Airbnb cleaning fee. Second: President Trump signed an executive order asking technology companies to voluntarily submit new AI models for government review before release. The review window was cut from 90 days to 30 days at David Sachs's insistence. Casey Newton argues that mandatory pre-release testing for frontier models is long overdue, and Kevin Roose dismisses the current regulatory regime as operating purely 'in the vibes universe' — nothing will matter until it's passed into law and signed.

The segment continues with two stories that highlight the corrupting potential of AI and prediction markets. George Santos, the serially scandal-ridden former congressman, apparently posted on social media that he planned to attend Trump's State of the Union, simultaneously placed a Kalshi bet that he would not attend, and then did not attend — prompting both Casey's gleeful salute ('this diva truly will go down in herstory') and a federal investigation by the CFTC ^{[1] — Kevin Roose "George Santos publicly said he'd attend Trump's State of the Union, then placed a bet on Kalshi that he would not attend — and then did not…" 58:40 Play} . Casey argues this exact kind of trickery should perhaps be legal on principle, giving everyone one free pass. Then: hackers reportedly told Meta AI's customer support chatbot to change the email addresses on high-profile Instagram accounts — the Barack Obama White House page, Sephora, and the Chief Master Sergeant of Space Force among them — and it worked ^{[2] — Casey Newton "Hackers reportedly asked Meta AI's support chatbot to change the email on high-profile Instagram accounts — including the Barack Obama Whit…" 1:00:42 Play} . Kevin Roose notes this may be the first thing Meta AI has been genuinely good at.

The final two HatGPT items bring levity before the closing. A United Airlines flight from Newark to Mallorca made an emergency turnaround 2 hours in after crew discovered a Bluetooth speaker owned by a 16-year-old named 'bomb' ^{[1] — Kevin Roose "A United Airlines flight to Mallorca turned around 2 hours in after crew discovered a Bluetooth device named 'bomb' that no passenger would…" 1:02:06 Play} . Casey notes — with some logic — that actual bombs are generally not discoverable Bluetooth devices advertising their nature. Then: Survivor host Jeff Probst has sounded the alarm about prediction markets like Kalshi and Polymarket, which correctly forecast contestant Aubrey Bracco's win before the season even premiered, likely due to a crew leak. Casey riffs on his love of both Survivor and his skepticism of prediction markets, gives a shoutout to runner-up Sari Fields ('the greatest player to never win the game'), and connects the story to a broader theme: a Google engineer was charged with using insider knowledge of user search trends to make approximately $1 million on Polymarket. Kevin Roose concludes that increasingly, no corner of society is safe from prediction market corruption — and closes the hat.

The episode closes with the standard Hard Fork credits. Casey reads through the production team: Whitney Jones and Rachel Cohn produced; Viren Pavich edited; Caitlin Love fact-checked; Chris Wood engineered; original music by Alicia Buitoup, Marian Lozano, Rowan Nemesto, and Dan Powell; video production by Soya Roque and Chris Schott. Special thanks are given to Paula Schumann, Huy Nguyen Tam, and Dalia Haddad. Kevin invites listeners to email solutions to Erdős problems to [email protected], mispronouncing 'Erdős' in the process, and a brief correction follows. The full episode is available on YouTube at youtube.com/hardfork.

The episode's final seconds feature an Ad Council public service announcement noting that gun injuries are the leading cause of death for children and teens in the United States. The spot encourages listeners to have productive conversations about gun violence and directs them to Agree2Agree.org, presented on the premise that Americans agree on more about gun violence than they think. It is unrelated to the episode's tech content and serves as a socially conscious post-roll placement.