Stripe confirms the OpenRouter deal + record AI concern

Stripe has confirmed the acquisition of OpenRouter, the neutral routing layer that sits between developers and the model providers. Sources told the New York Times the deal values OpenRouter at $7.5 billion, roughly six times the $1.3 billion it carried in May; founders receive about $1.5 billion and investors about $6 billion. OpenRouter says its mission, name, product and roadmap are unchanged, and that routing "doesn't bend to any model, any provider, or any parent company." The transaction is expected to close in the coming weeks.

Pew Research Center surveyed 3,488 US adults in late June and found 52 percent are more concerned than excited about the use of AI in daily life, against just 9 percent more excited — up from 37 percent concerned in 2021. For the first time a majority of adults under 30, at 55 percent, sit on the concerned side, and 71 percent now expect AI to mean fewer jobs over the next twenty years, up from 64 percent in 2024. The industry's working assumption that familiarity would breed acceptance is not holding.

Ornith-1.5 arrived as three open-weight models — a 397 billion parameter mixture-of-experts (MoE) flagship, a 35 billion parameter MoE, and a 9 billion parameter dense model — all under a plain MIT license, with the weights already downloadable. The flagship scores 86.1 on Terminal-Bench 2.1 against Claude Opus 4.8's 85.0, and 56.0 on DeepSWE against 59.0. Two things make it unusual. The training pipeline has the model propose its own tasks and build its own scaffolds, and there is no hosted API at all, so running the flagship means eight H200 cards of your own.

WIRED found that Meta ran 32 advertisements for Kromix, an app marketed as an "AI image styler," in which a woman resembling a prominent US politician appeared before the spot cut to pornographic video. The campaign was targeted at men only and sold the tool as having "no restrictions." Apple removed Kromix from the App Store, saying the app "added prohibited features" after review and that nudification apps break its guidelines. Meta pulled the campaign only after WIRED asked about it.

OpenAI is previewing Private Safety Processing, which watches API inputs and outputs for abuse that spans several conversations — the case where malware engineering is split across separate requests — while retaining none of the customer's content. When it fires, only a narrowly defined signal about the type of activity reaches OpenAI, which then asks the customer for context before any enforcement. The pitch is aimed at Anthropic, whose July policy holds session data for 30 days on covered models such as Fable, a retention window that enterprises handling sensitive material have objected to.

TerraPower's Natrium design pairs a 345-megawatt sodium-cooled fast reactor with gigawatt-scale molten salt storage, so the reactor never throttles down: surplus heat goes into the salt and comes back as extra steam when a data center's load spikes. The first plant is under construction in Wyoming under a Department of Energy cost-share worth up to $2 billion, and a second breaks ground in 2027 for an unnamed data center customer. Meta agreed in January to buy eight Natrium plants. The storage was designed for intermittent renewables rather than for AI; the fit with data center load was a happy accident.

Google added a batch of study features across Search and Gemini: AI-generated interactive visuals and 3D simulations, custom practice quizzes, full-length SAT practice tests, a Lens mode that explains a photographed worksheet, and a student hub holding notebooks, flashcards and quizzes. The sharper move is commercial. Eligible US college students get a year of Google AI Pro at no cost, a plan that normally runs $200 a year, and it auto-renews at full price when the year ends. Students outside the US get the cheaper AI Plus tier instead.