Anthropic's Tom Brown ended the June model-safety standoff
Two stories today that have nothing to do with benchmarks: how one lab actually resolves a fight with Washington, and what publishers do with AI when nobody is watching.
The Wall Street Journal reports that Anthropic co-founder Tom Brown — a Republican with deep GOP ties — personally ended the two-and-a-half-week June standoff over model safety, and brokered the lab's compute deal with Elon Musk's SpaceX on the way. According to the Journal's profile, Brown's Washington relationships were the lever that unfroze Anthropic's Fable 5 and Mythos 5 models: Commerce Secretary Howard Lutnick's June 30 letter lifting the restrictions was addressed to Brown directly, and officials who had been sitting across from CEO Dario Amodei moved to Brown in those meetings — earlier reporting put it bluntly, they simply liked him better personally. The SpaceX side of the ledger is now concrete: Anthropic agreed to pay $1.25 billion per month through May 2029 for capacity on the Colossus 1 cluster, with a discounted ramp in May and June 2026, a 90-day termination clause for either side, and a term total above $40 billion, a tally Anthropic's spokesperson confirmed to Business Insider. Two caveats keep us honest: the GOP-ties framing is the Journal's own, not independently corroborated, and the sequencing is easy to blur — the Musk deal was announced in May, before the standoff, and the standoff was with the White House, not Musk. Why it matters: compute access and regulatory goodwill have quietly become the same currency, and Anthropic's answer to both was to put a dealmaker who isn't the CEO in the room. We tracked the SpaceX cloud pivot when it broke — SpaceX's AI unit turns itself into an AI cloud firm — and this is the political machinery behind it.
Three of the Big Five US book publishers — HarperCollins, Simon & Schuster and Hachette — are quietly using AI to write back-cover copy, generate cover art, run publicity, and even email literary agents, often at executives' direction and without author consent, a WIRED investigation finds. The reporting rests on interviews with more than two dozen staff: HarperCollins bought Claude licenses from Anthropic and assigned dozens of employees as volunteer "AI Champions," plus ChatGPT and Jasper seats, while Simon & Schuster ran an internal AI-use contest with a $10,000 grand prize. The revolt half of the story checks out independently: Simon & Schuster staff are circulating an open letter to CEO Greg Greeley over a trial of Skan AI's Blueprint monitoring software, which Publishers Lunch and NY Magazine's Book Gossip both reported before WIRED landed — Greeley's memo says no decision has been made, which denies the monitoring, not the AI-written copy. Hachette said it supports AI for operational purposes but "does not support creative uses of AI, including communicating with our authors"; Simon & Schuster called its use limited and non-mandatory; HarperCollins didn't respond. The copy-and-cover specifics currently rest on WIRED's anonymous sourcing alone — but the contradiction is already there: the houses that have spent two years objecting to how models were trained on their books are now using those same models on their own product pages, a tension we've examined in Training AI on books is fair use — piracy is the crime.
What to watch: whether any publisher puts AI use into book credits or author contracts before trade press forces the question.
Should publishers be required to disclose AI use to authors — and to readers? Tell us in the comments.
