On June 12, Commerce Secretary Howard Lutnick sent Anthropic a letter, and the company's two most capable models — Fable 5 and the restricted Mythos 5 — went dark for users worldwide. The order, reported to cite the Export Control Reform Act of 2018, barred any foreign national from accessing the models. Rather than try to verify the nationality of millions of users in real time, Anthropic pulled both offline and left its other models running.

Joel and Stephen, calling in from Sydney, trace how it happened: a launch on June 9, a CEO essay two days earlier asking Washington for the authority to block unsafe models, and a reported bypass found by an Amazon-linked partner that reached senior officials within days. From there they get into what it means to build a business on a model a government can switch off overnight, why Europe read the whole thing as an argument for its own sovereign AI, and whether Dario Amodei calling the directive a "misunderstanding" is a principled stand or the regulatory-capture critique his rivals keep making.

Also in this one: Sam Altman's bomb-shelter line, the case for proprietary models over frontier dependency, Anthropic's quiet IPO filing, and why both hosts think a government in "freak out mode" may have to walk this back. Recorded Tuesday, June 23 — with the directive still standing and G7 fallout still landing.

Joel and Stephen open on Anthropic's June 4 blog post "When AI Builds Itself," which calls for the industry to keep open the option of pausing frontier AI development — conditional on rival labs verifiably doing the same. The timing is the story: Anthropic confidentially filed an S-1 with the SEC on June 1, reportedly targeting a valuation between $965 billion and $1 trillion on roughly a tenth of that in revenue. With Rob Enderle of the Enderle Group calling the post strategic marketing rather than a concrete initiative, the hosts work through whether an industry-wide pause is even enforceable when Chinese labs aren't slowing down, and what a Walmart-sized valuation gap says about where the AI market sits right now.

The second segment turns to Apple's N50 smart glasses, which Bloomberg's Mark Gurman now reports have slipped from late 2026 to late 2027. The first generation will ship with cameras, speakers, and Apple Intelligence but no in-lens display, priced between $200 and $500 against Meta's Ray-Bans — which currently move about 85% of global AI-glasses volume. The project lands squarely on Tim Cook's desk before he hands the CEO role to John Ternus on September 1. Stephen makes the case that Apple is chasing visual AI and tight iPhone integration rather than a Meta clone, with the caveat that the voice assistant needs a real upgrade before any of it works.

Then the AI layoff reversal. A February CareerMinds survey of 600 HR professionals found two-thirds of employers who ran AI-driven layoffs are already rehiring, and over 90% regret the cuts. Gartner projects half of all AI-attributed customer service reductions will be undone by 2027. Klarna CEO Sebastian Siemiatkowski, who once said AI was doing the work of 700 customer service agents, admitted the company "went too far" — even as Intuit is still cutting 3,000 jobs this summer citing AI. Joel and Stephen argue the layoffs were rooted in profitability and geographic labor arbitrage more than AI itself, and flag a separate reckoning coming once companies see their first real token bills.

The episode closes on a Stanford, Chapman, and Northeastern study of the Pymetrics hiring platform — the largest independent audit of AI hiring algorithms to date, covering over 4 million applications between 2018 and 2022 and slated for the ACM FAccT conference in Montreal. The researchers found that more than one in four Black applicants went to roles where the algorithm produced outcomes severe enough to trigger federal discrimination scrutiny, and because scores are reused across employers, a rejection by one company predicts rejection by the next. With the EU AI Act now classifying hiring algorithms as high-risk under Annex III, the conversation lands on whether a score generated in January should still be deciding interviews in June.

In April 2026, Elon Musk and Sam Altman faced each other in a federal courthouse in Oakland, California — two men who co-founded OpenAI in 2015 on a shared promise to build AI for humanity, now arguing in front of nine jurors about who owns the soul of it.

In this episode, Joel and Stephen go deep on the full story. Who are these two men, really? Where did they come from, what drives them, and were they ever actually compatible? Joel breaks down the founding, the power struggle, and the trial itself — including the Greg Brockman diary entry that's now sitting in front of a federal jury, and the Microsoft partnership restructuring that OpenAI announced on the same morning the trial opened. Stephen, drawing on decades of experience inside technology organisations, gives his unfiltered read on whether this is a values clash, a power struggle, or — as he puts it — just petty bullshit at the highest level of the industry.

They also get into what the verdict actually means: whether OpenAI is already too big to unwind regardless of the outcome, why Anthropic may be the quiet winner of this whole saga, and what it signals about AI governance when the most important question in the industry is being decided by a judge in Oakland rather than anyone who was elected to decide it.

Stephen's prediction: OpenAI wins. Musk walks away looking like an angry, disgruntled founder. The law book will settle it — and it'll come down to what was written, not what was promised over dinner in 2015.

Also referenced in this episode: Van Wijk & Ferreira Gomes, "The GOALS of the Techno-Libertarian", Netherlands Institute of International Relations, 2026 · Drexel & Withers, "Terms & Concerns," Catalyzing Crisis, Center for a New American Security, 2024 · Fuchs, "The World in the Age of Trump 2.0", University of Westminster Press, 2025 · Polan, "Growth's Imagination", Bristol University Press, 2025

Signal — listen wherever you get your podcasts.