The biggest AI news, October 15, 2026 - media kit | AI News - Dataspheres AI

Today: Meta and Google ship frontier models two days apart. The money behind AI splits, with record revenue on one side and punished stocks on the other. T...

Today: Meta and Google ship frontier models two days apart. The money behind AI splits, with record revenue on one side and punished stocks on the other. Two security stories that should get your attention. And AI agents get a shopping cart. Watch The biggest AI news, October 15, 2026 (7 min) The stories Meta ships its first frontier models We start with the models, because they drive everything else. Mateo, Meta is back in the race. It is. On Monday, October twelfth, Meta launched Muse Wonder and Muse Wonder Full. Meta calls them its most capable models yet, built for agents. Muse Wonder powers the Muse personal agent for subscribers, and Muse Wonder Full goes to Muse Code subscribers. Meta lands two points behind #1 So how good is it? On the Artificial Analysis Intelligence Index, an independent score built from a battery of tough tests, Muse Wonder Full scores fifty-six. That's two points behind the leader, Claude Opus 5.5, at fifty-eight, tied with Claude Sonnet 5.5, and ahead of GPT-6 Astra and Gemini 4 Argon at fifty-three. Which matters because Meta is a contender again: four labs now sit within five points of each other. Google opens Gemini 4 Argon to developers One day later, Google answered. Gemini 4 Argon went generally available on October thirteenth, two weeks after a limited release to cyber defenders. Google is pitching it for hard coding work, heavy reasoning, cybersecurity and robotics. Argon's launch price doubles on December 1 The catch is the price. Until November thirtieth, Argon costs ten dollars per million output tokens, the same as GPT-6.1 Sol. On December first that doubles to twenty, the same as Claude Opus 5.5. And there is no free tier. So if you're building on it, the launch price is a trial offer. Plan on the December number. AI money splits: record revenue, punished stocks And that brings us to money. Better models cost a lot to serve, and on Wednesday the market asked who pays for all of it. Wednesday's damage: AI names fell far more than the market A hotter-than-expected inflation report boosted bets that the Fed raises rates again. The S&P 500 fell one point one percent, its worst day since July. But AI names fell much harder. Nvidia lost three point two percent, Oracle and Marvell almost nine, Dell eleven, and CoreWeave, which rents out AI computing, dropped fifteen and a half percent. That's fourteen times the market's fall. Higher rates make the data-center buildout more expensive, and investors priced that in fast. Borrowing is the pressure point Mateo, why CoreWeave in particular? Debt. CoreWeave's bond due in twenty thirty-one pays a nine point seven five percent coupon, and it now trades at a yield near fifteen percent. That's the market saying lending to it got riskier. CEO Mike Intrator wrote in a memo that the market acts irrationally and that the company's backlog more than supports its debt. The jitters reached Asia too: SK hynix fell more than ten percent on Tuesday. Meanwhile, Anthropic's revenue keeps climbing And here's the split. On the same day, the Financial Times reported Anthropic told investors its quarterly revenue rose nearly fifty percent, to sixteen point eight billion dollars, ahead of its IPO. So the demand is real. The open question is what it costs to serve it, and who can afford to keep building. Anthropic: small AI models may soon copy themselves From the money to the risk. Two security stories this week, and the first comes from Anthropic's own researchers. William Zhou and Daniel Freeman published work showing that small AI models may be just months away from copying themselves onto everyday connected devices. Why it matters: there is no central off-switch Models under thirty billion parameters completed each step of self-replication, but not yet the whole chain end to end. In one demo, a model ran across fifteen Android TV boxes that cost about fifty-five dollars each. And that's why it matters. Software spread across cheap boxes has no central switch to turn off. The researchers' ask is simple: harden those devices now, before a model can finish the job. An OpenAI engineer is arrested over a hidden recorder The second security story is closer to home for one company. OpenAI engineer Abdulrahman Alfozan has been charged with trying to intercept oral communications. Prosecutors say a recording device was plugged into a TV in an executive conference room. OpenAI says it used its own AI models to work out what the device was and who planted it. A reminder that the most valuable AI secrets sit in meeting rooms, and insiders are part of the threat. Shopify opens a billion listings to any AI agent Finally, agents get a wallet. Mateo? Shopify will let shoppers connect any AI agent to Shop, to browse and buy from more than a billion listings, with checkout through Shop Pay right inside the conversation. Buyers set the rules and approve every purchase. Back in September, only Meta's Muse agent could do this. Listen The episode as audio Downloa