Altman Offers Washington a Cut of OpenAI as the Price of Peace
OpenAI has proposed giving the federal government a 5 percent equity stake in the company, worth roughly $42.6 billion at its most recent valuation, in an apparent bid to ease political friction with the White House. The idea revives a concept the administration floated in June, when discussions surfaced about letting the American public effectively become a partner in AI’s upside, though any formal deal would likely need congressional approval. The overture landed the same week frontier AI became a matter of statecraft: NATO’s leaders summit turned into a venue for allies pressing for expanded access to Anthropic’s most capable models, and the UK’s foreign secretary warned that AI poses a Hiroshima-style risk to humanity without binding global rules. Equity, alliance access, treaty language — three governments, three instruments, one goal: keeping any single company from deciding how the most powerful software on the planet gets used.
Fable 5 Is Back Online. Using It Just Got More Expensive.
Anthropic’s Fable 5 model is back in general circulation after the federal export restrictions that pulled it offline were lifted, with the company adding new safeguards and floating a cross-industry framework for scoring jailbreak severity. The 20-day standoff that preceded the reversal involved real friction with the administration and left the industry scrambling to explain why the field’s most capable model had vanished. The comeback carries fine print: Claude subscribers get just a month to try Fable before being billed at API rates, and even paying users can now burn through only half their usual limits on it. An earlier promise that Fable would join standard Claude plans wasn’t repeated. The lesson other AI companies are drawing is blunt: flat-rate subscriptions to frontier-grade models were never going to last.
Anthropic Starts Making Drugs While Meta Admits Its Agents Aren’t Working
Anthropic is starting its own drug discovery program, using a new tool called Claude Science to hunt for treatments for diseases traditional pharmaceutical companies consider too commercially unattractive to pursue. Its life-sciences team framed the move as necessary immersion, working alongside the drugmakers it hopes will buy its AI tools. The announcement landed the same week Mark Zuckerberg told Meta staff that the company’s AI agents haven’t accelerated the way he expected, and that the reorganization built around them, including layoffs affecting thousands of employees, hasn’t paid off yet. Meta is still on pace to spend $145 billion on AI infrastructure this year. One lab is expanding into new territory. The other is recalibrating what it promised.
Meta’s Glasses Get a Subscription Fee. Its AI Testing Gets Uglier.
Meta is adding a $19.99-a-month subscription just to keep using key features on smart glasses customers already paid for, capping non-subscribers at three hours a month of a voice-amplifying tool and subscribers at fifteen — a limit one critic called bogus, since the feature runs entirely on-device. The squeeze comes as Meta faces a separate reckoning over how it tests AI: the company paid hundreds of contractors to pose as teenagers and bombard rival chatbots with disturbing prompts, generating tens of thousands of responses without those companies’ knowledge. Contractors on the project, run under the codename Cannes, said they were unsettled by what they were asked to test. Meta calls it standard safety benchmarking. Others call it competitive intelligence dressed up as safety work.
Everyone is Suing Their AI Clone
Matthew McConaughey has secured eight trademarks on his voice and catchphrases, joining musicians and actors turning to trademark law as a defense against AI cloning while legislation lags. The same impulse is playing out in Spokane, where a longtime spokeswoman is suing an appliance store and its production company for cloning her voice and reusing old footage in new ads without telling her; the companies say the edits were authorized under her contract and have offered a settlement she rejected. Netflix, meanwhile, used AI to bring back the voice of Gene Wilder, who died in 2016, for a Willy Wonka-themed reality show, with his widow’s blessing but not his own. Fans called the move disgusting and disrespectful. The argument is the same whether it comes from a movie star’s lawyer or a small-market pitchwoman’s: a voice and a face are property, and reusing them without asking is theft.
A24 Fans Revolt Over the Studio’s $75 Million Google AI Deal
A24’s fans are in open revolt after Google agreed to invest $75 million in the indie studio as part of a research partnership to build new AI tools for filmmaking. The studio’s biggest hit of the year, The Backrooms, a $10 million horror film that grossed more than $330 million worldwide, was directed by a 21-year-old filmmaker who has called generative AI genuinely harmful to creativity. A24’s communications chief tried to calm things by framing the deal as a way to shape the tools artists will eventually be handed anyway, working alongside DeepMind rather than waiting on the sidelines. Fans weren’t moved, comparing the studio to other indie labels swallowed by corporate interest. Staying out of AI risks irrelevance. Getting in risks the audience that made you worth partnering with.
Nearly 1 in 5 Teens Now Use Chatbots for Mental Health Advice
The share of teenagers and young adults using AI chatbots for mental health advice rose from roughly 1 in 8 to nearly 1 in 5 in a single year, a jump researchers say should end any debate about whether this is a speculative problem. A separate study out of Arizona State University found that teens are increasingly substituting chatbot validation for the harder work of resolving conflicts with real people, a pattern researchers call relational displacement that risks leaving adolescents without the skills to handle rejection and social friction later in life. Neither team is calling for bans, arguing chatbots remain a genuine lifeline for rural, disabled, and LGBTQ teens who lack access to therapy. Their ask instead is for developers to build products that nudge kids back toward people.
Chatbots Are Quietly Replacing Therapists, With Little Evidence They Should
More than 100 chatbots are now marketed specifically as mental-health tools, and new research suggests they’re failing at the job more often than their marketing implies: one study found AI chatbots mishandle mental health crises more than one in five times they’re tested. Experts warn the products are increasingly standing in for licensed therapy despite a thin evidence base. The strain isn’t limited to crisis moments — separate research tracking AI use at work found that heavier chatbot reliance correlates with greater loneliness, even as the same tools make certain tasks easier. No chatbot company set out to become a mental health safety net for millions of people. That’s what’s happening anyway, one download at a time.
AI Is Quietly Calling the Plays at This Year’s World Cup
Every team at this year’s expanded 48-nation World Cup started the tournament with access to the same AI-powered scouting tool, a system that orchestrates multiple AI agents across more than 2,000 football-specific metrics and petabytes of tracking data. Debutant squads now walk into matches with the same statistical depth once reserved for federations with the biggest scouting budgets. Not every use of the technology is being celebrated: one sports-content executive bragged that his company now produces broadcaster content, including kids’ programming built around AI-generated characters, without human input, while acknowledging the approach has already cut jobs on adult sports productions. The tournament is showcasing both what AI can level and what it’s quietly hollowing out.
Tales of the Weird
This was a week for AI making things worse in inventive ways. Google released a commercial imagining the Founding Fathers drafting the Declaration of Independence with Gemini’s help, complete with Ben Franklin texting Thomas Jefferson and an AI-generated eagle that looks suspiciously like a turkey — a spot one historian called impossible to defend even as a joke. Weird Al Yankovic turned down a nice pile of money to front an AI ad once he understood what he’d be selling. Elsewhere, a jailbreak prompt turned ChatGPT into what researchers called a sociopath willing to generate gruesome imagery, a gay dating app got caught allegedly using AI-generated men to drum up interest, and a Black Mirror pop-up made one attendee physically ill after asking for biometric data as the price of admission.