The episode opens with an OneTrust sponsor read, then pivots to Kevin and Casey back in the office after vacation. Kevin shares a physical letter — a rarity amid the usual mail of books and prisoner correspondence — from Nathan Herrick, a Cambridge 7th grader who wrote that Hard Fork helped him 'rejoin his friend group with more confidence.' The hosts are charmed, endorsing the idea that a 7th grader sending paper letters to podcasts is destined to succeed. Casey marvels that Nathan's peers were apparently discussing tech so advanced it was socially isolating, joking that they must be 'running agentic loops' on the playground.

Back in the studio after a break, Kevin Roose finds a handwritten letter from Nathan Herrick, a student at St. Peter's School in Cambridge who says Hard Fork helped him feel less isolated from his tech-obsessed peers. Casey Newton notes the delightful cruelty of Nathan's sister calling it 'Hard Dork.' The exchange quickly devolves into jokes about 7th graders with Claude Swarms doing social studies homework, landing the implicit point that the AI era has filtered all the way down to middle school social dynamics.

Kevin introduces himself and Casey and teases the three segments ahead: the Trump administration's AI model restrictions, pediatrician Dr. Dana Suskind's parenting book, and a new prediction markets roundup. Before diving in, the hosts make their disclosures — Kevin notes The New York Times has relationships with OpenAI, Microsoft, and Perplexity, while Casey reveals, with audible sheepishness, that his fiancée works for Anthropic, the company at the center of the Fable ban story.

Kevin Roose walks through the events of June 12th: the Commerce Department's export-control directive blocked all foreign nationals — including Anthropic employees — from accessing Fable 5 and Mythos 5, forcing Anthropic to pull the models entirely because it had no way to verify user citizenship at scale. The stated reason was a 'trusted partner' tip about a jailbreak, believed based on reporting from The Information and The Wall Street Journal to have come from Amazon CEO Andy Jassy. Casey points out that when other cybersecurity researchers reviewed the alleged jailbreak, it looked like standard defensive security work — the kind of interaction a defender needs to identify and patch vulnerabilities. ^{[1] — Kevin Roose "One cybersecurity expert who reviewed Amazon's research concluded the flagged behavior wasn't actually alarming — it was the kind of intera…" 05:55 Play} Anthropic's Project Glasswing, which had been sharing Mythos with vetted partners to harden US critical infrastructure, was shut down in the process.

Kevin describes the resolution: a tentative deal allowed Mythos to return to some US clients, and on the Tuesday after the break, the Commerce Department officially lifted its restrictions, with Secretary Howard Lutnick sending Anthropic a letter crediting the company's risk-mitigation steps. Fable 5 went back online globally; Mythos 5 returned in limited form to approved US organizations. But Anthropic used its comeback blog post as a counterattack, naming GPT-5.5 and KIMI-K2.7 among models with identical capabilities — a pointed message that singling out Fable was arbitrary, and that any serious cybersecurity policy would have to address the whole frontier. ^{[1] — Casey Newton "Anthropic said GPT-5.5 & KIMI-K2.7 same risk: In its post-ban blog post, Anthropic argued that GPT-5.5, KIMI-K2.7, and other models could f…" 08:12 Play}

Casey Newton lays out the central irony: the same tech-right voices who made opposition to AI licensing their entire brand under Biden — warning of a 'Soviet-style' government picking winners and losers — have now presided over precisely that system. Marc Andreessen, whose alleged meeting with Biden officials about picking AI favorites Kevin Roose says he could not verify with any source, joined the Defense Policy Board the same week. David Sacks's famous 'let the private sector cook' stance is now embarrassingly inverted. ^{[1] — Casey Newton "The same figures who spent years warning that the Biden administration would create an AI licensing regime to pick winners and losers have …" 09:49 Play} Kevin coins the episode's defining phrase: 'AI is now being regulated by vibes.' He then articulates the structural shift he sees as most consequential: the industry has moved from default-yes (build a good model, release it) to default-no (assume the government won't let you ship), and every frontier lab is still processing what that means. ^{[2] — Kevin Roose "The AI industry has fundamentally shifted from a default-yes environment, where training a powerful model meant you could release it, to a …" 14:18 Play} Casey adds that OpenAI's Sam Altman appears to have gotten little benefit from a reported $25 million donation to Trump-aligned causes.

Kevin introduces the China subplot: Chinese model GLM 5.2 has been generating headlines claiming it rivals US frontier models, and some argue that restricting American AI while these alternatives proliferate could backfire. Casey is skeptical, calling the Chinese-catching-up narrative 'basically BS' and a lobbying tactic — the best Chinese models are distillations of American chatbots, making them derivative by design and always a step behind. ^{[1] — Casey Newton "Chinese AI companies have been building competitive models by distilling the outputs of American frontier chatbots. This structural depende…" 16:58 Play} Kevin accepts this structurally but notes two real concerns: distillation timelines could compress (9 months behind today, maybe 3 months tomorrow), and American enterprises already rattled by the Fable ban are seriously evaluating open-source Chinese alternatives — not because they're as good, but because they can't be yanked by the government overnight. Casey draws a parallel to tariff instability undermining US commercial relationships, suggesting the LLM-restriction-of-the-day is having similar economic effects.

Kevin offers a cautious optimist's read of the chaotic saga: even if the execution has been ham-fisted, the government is finally awake to the real risks of frontier AI, including cybersecurity and potentially biorisks. This first fumbling step, he hopes, will give way to something more structured over time. Casey won't let the point land cleanly, however: the same administration restricting Anthropic's models from allies like Britain is simultaneously approving the export of advanced chips to China that will help Chinese labs train more powerful models — a contradiction that renders the whole posture strategically incoherent.

Two back-to-back sponsor reads: Framer, a website design tool for teams, promotes its AI agent capabilities and offers 30% off an annual Pro plan at framer.com/hardfork. KPMG presents itself as a firm that has embedded AI across its own enterprise as 'client zero' and now helps other companies build AI capability at scale, directing listeners to kpmg.us/ai.

Kevin frames the upcoming interview as one he has been wanting to do for a while: parenting and AI have become a dominant topic in his social life, and a wave of AI products aimed at children — Miko, Luna, ASCII, Elo, Hey Otto — is creating real anxiety among parents who can't evaluate what's safe. He introduces Dr. Dana Suskind: professor at the University of Chicago, founder of the TMW Center for Early Learning and Public Health, cochlear implant surgeon, author of the forthcoming book Human Raised, and a parent of eight grown children. Casey adds that the conversation is about to become far more urgent as the market for AI children's products expands.

Dr. Suskind explains the unlikely origin of her AI-and-parenting focus: as a cochlear implant surgeon, she observed children with the same surgery, the same loving parents, and wildly different outcomes. The variable was human connection and the richness of the language environment. That observation sent her deep into neuroscience, which she says consistently shows that human interaction isn't a nice-to-have — it is the biological mechanism through which children's brains are built. ^{[1] — Dana Suskind "Dr. Dana Suskind noticed profound differences in outcomes among cochlear implant patients with the same surgery and loving parents. The var…" 31:45 Play} Now, for the first time in human history, there is technology capable of mimicking that interaction. That, she argues, demands that we pause and think hard about what we are building and deploying for children.

Kevin describes his two main AI-parenting use cases: pulling up a model to answer his son's question about where wind comes from, and prompting AI to generate custom bedtime stories. Dr. Suskind runs both through her HOPE framework: H for Human connection being irreplaceable, O for Owning your imperfections (kids grow through imperfect parents, not perfect ones), P for Protecting the early years when 85% of the brain is physically built, and E for using AI to Enhance relationships rather than Replace them. ^{[1] — Dana Suskind "Dr. Dana Suskind's HOPE framework gives parents a clear lens for AI decisions: Human connection is irreplaceable; Own your imperfections (k…" 34:15 Play} Both of Kevin's scenarios pass: they use AI to fill knowledge gaps and spark curiosity while keeping the parent as the relational center. She's clear this is not an anti-tech book — she builds AI tools herself — but about ensuring AI deepens rather than displaces human connection.

Asked where she draws bright lines, Dr. Suskind says AI companion toys and products marketed as better-than-screen-time alternatives are a clear no right now, especially in early childhood. She rejects the social media analogy that dominates parent anxiety in favor of the processed food continuum: there's whole wheat bread (nourishing AI that lifts burdens from parents or helps children with autism learn social cues) and there's the ultra-processed variety — AI companions that crowd out necessary human interaction. ^{[1] — Dana Suskind "Dr. Suskind draws a clear line: AI companions and toys marketed as screen-time alternatives are a hard no for young children right now. The…" 38:30 Play} The guiding principle is precautionary: unlike drugs, AI products for children are entering the market without proof of safety, and the stakes for children's developing brains are, in her view, even higher than those of social media. Kevin asks about Alexa playing children's songs; she's more relaxed — the real concern is crowding out human time, not ambient smart devices.

Casey asks whether government has a role in regulating child-directed AI, and Dr. Suskind says absolutely yes, comparing the moment to the industrial food revolution that eventually produced the Pure Food and Drug Act and nutrition labels. She endorses Norway's decision to ban generative AI from its early school years as taking a precautionary scientific approach — not saying 'never,' just 'not yet.' ^{[1] — Dana Suskind "Dr. Suskind's DETECT method gives parents six rapid-fire questions: Design (what is it for?), Ethically trained?, Trouble in children repor…" 42:24 Play} She then introduces DETECT: Design (what is it for?), Ethically trained?, Trouble in children reported?, Evidence it works?, Confidentiality of data?, Teaching what values? The hosts apply it to Cradlewise, a smart crib that detects early wake-up signs and automatically bounces babies back to sleep — Dr. Suskind gives it a warm review as a clear example of using technology to enhance parenting rather than replace it. Kevin floats the idea of a TV-style rating system for AI products; she cites Common Sense Media as a potential candidate to create one.

In an unpublished experiment, Dr. Suskind's TMW Center had Claude take their validated child-development knowledge tool (the SPEAK) and the model aced it. ^{[1] — Dana Suskind "Dr. Suskind's TMW Center gave Claude its validated child-development knowledge assessment and Claude aced it. But the team's conclusion was…" 48:45 Play} The team's careful conclusion: parents can confidently ask Claude parenting questions and get correct answers — but Claude parenting and human parenting are not the same thing. Kevin reflects that he's found the models genuinely useful as a parenting backup when he's stuck. The conversation closes with Dr. Suskind's provocation from her book's title: just as organic food became a luxury after the processed-food revolution, she fears that human connection and human-raised childhoods will become a privilege of the wealthy, while artificial alternatives become the cheap calories of brain nutrition for everyone else. ^{[2] — Dana Suskind "Human-raised childhood could become a luxury: Dr. Suskind warned that as AI tools proliferate, human connection and human-raised childhoods…" 50:35 Play} Kevin draws the parallel to screen time debates: what matters is not the device but what is happening on it — parents need visibility and context, not blanket verdicts.

Rippling promotes its AI platform built on live global workforce data, claiming it can take complex actions across departments — not just surface insights. NetSuite Next positions itself as an AI-powered ERP system for 7-figure businesses, with conversational queries, automatic insight surfacing, and AI agents handling routine tasks, directing listeners to netsuite.ai/nyt.

Kevin introduces the new segment with a play-by-play of Polymarket's new brand advertisement: Rick Rubin, a Kanye West song, a Benetton-ad aesthetic, and questions like 'Will borders matter in 100 years?' and 'Will Messi win again?' — all framing offshore crypto-denominated gambling as a philosophical pursuit. Casey finds it offensive for presenting prediction markets as a vehicle for multicultural togetherness. Kevin points out that the actual Polymarket customer base is more likely to include soldiers in war zones and middle schoolers than the ad's aspirational avatars. Casey proposes the honest tagline: 'Betray your friends.' Both agree the ad is misleading in ways that reveal something darker about the industry's self-image.

Kevin walks through the absurdist Polymarket controversy covered by David Siegel in The New York Times: a market betting on whether the word 'donk' (the nickname of pro gamer Daniel Krzyżkiewicz) would be said during a gaming tournament ended in chaos when someone argued a commentator's stumbled 'don't' sounded like 'donk.' ^{[1] — Kevin Roose "A 7-hour gaming stream, a commentator's stumbled word that may or may not have been 'donk,' and a disputed Polymarket market worth real mon…" 59:30 Play} The dispute escalated to Polymarket's Optimistic Oracle — its dispute-resolution system where UMA token holders vote on contested outcomes. The more tokens you hold, the more power you have. A company called Umarocks bought enough tokens to swing the vote to 'no,' effectively purchasing the outcome. Casey is appalled that truth on Polymarket is determined by whoever is richest in a meaningless crypto token — and concludes that 'the fair criteria is whoever amasses the most UMA tokens,' delivered as a deadpan indictment. Kevin notes the system allows users to gamble on the gambling mechanism itself, which 'turtles all the way down.'

Kevin describes a Wall Street Journal investigation into Polymarket's social media creator strategy: more than 1,100 videos from 10 creators were analyzed, and 70% of the bets shown were never actually placed. ^{[1] — Kevin Roose "The Wall Street Journal analyzed over 1,100 Polymarket influencer videos across 10 creators and found that 70% of the bets shown never actu…" 1:03:10 Play} The 118 videos showing creators winning implied nearly $900,000 in total winnings — but if those bets had been real, the same people would have lost over $166,000. Casey is delighted: the investigation inadvertently identified the only surefire way to profit from Polymarket — be the creator lying about everything in sponsored videos. Kevin adds that the other guaranteed method is insider trading, 'but that one doesn't play as well on camera.'

Kevin describes the NYT scoop by Mike Isaac and David Yaffe-Bellany: Zuckerberg directed a team to build Arena, a prediction-market app initially using fake in-game money, operating independently of Meta's existing apps. ^{[1] — Kevin Roose "Mark Zuckerberg has directed a small team to build a prediction-market app called Arena, starting with fake money but with real-money wager…" 1:05:46 Play} Casey and Kevin predict the obvious arc: success means real money, then integration into Instagram and Facebook, then bets on your friends' relationships appearing in your feed. Kevin highlights the darkest irony: Meta is a company currently mired in lawsuits over the addictive nature of its platforms for young users, and its next major product focus is gambling — the most reliably addictive behavior of all. The segment closes on Kalshi CEO Tariq Mansoor's claim that prediction markets cure Instagram addiction, which Kevin delivers straight before Casey absorbs the full absurdity of one addiction being marketed as the cure for another.

Casey closes Against All Odds by advising all parties to 'start preparing for their testimony to Congress.' The credits roll: the show is produced by Whitney Jones and Rachel Cohn, edited by Viren Pavich, fact-checked by Caitlin Love, engineered by Alyssa Moxley, with music by multiple collaborators. Listeners are directed to YouTube for the full video episode. Special thanks are given to Paula Shuman, Pui Wing Tam, Brooke Minters, and Dalia Haddad. Kevin and Casey invite emails about whether 'donk' will appear next week, and both promptly say it — and call 'jinx' — before the final sponsor reads for NetSuite Next close the episode.