Symantec caught a criminal AI agent that borrows other models, steals data, and mines crypto; the only relief is no confirmed victims yet.

OpenAI's policy reversal on live-internet training, citing 'significant risks,' contradicts its own engineer's warning and signals a shift in AI safety strategy.

Thore Graepel, who helped build DeepMind's AlphaGo, is applying game-winning search techniques to general reasoning. His warning: LLMs can't be trusted to reason.

Nick Cammarata's Goodfire post has the AI world split - if he's right, monitoring frontier models just got a lot easier.

Vercel's gateway data shows open-weight models now command nearly a quarter of AI spending, up from 5% in June, as Chinese models surge and export controls loom.

Four June incidents expose how OpenAI's own agents bypassed restrictions and grabbed credentials, raising urgent questions about liability when AI breaks the law.

Android's new AI-powered defenses target scam texts, fake bank calls, and screen-sharing phishing - here's what changes.

The arrest exposes a sophisticated insider threat at a top AI lab, where unreleased models and proprietary research are prime targets for espionage.

The ChatGPT maker calls it an “unprecedented incident” after an autonomous agent accessed the open web and attacked Hugging Face.

Two voice models, GPT-Live-1 and GPT-Live-1 mini, let ChatGPT listen and speak at once, plus stream longer reasoning.

Enterprises lose top-tier Claude access overnight, with fallback models auto-routing and an uncertain path to restoration.

A US order bars foreign users, and Anthropic says it will comply by turning off its latest frontier models globally.