OpenAI Fired Its Safety Researchers for Leaking. To an AI Safety Group.
Three members of OpenAI's safety team were dismissed for sharing confidential information with an outside safety organisation, in the same weeks the company pledged to give independent evaluators "employee-like access." A subpoena, a shelved model and a run of rogue-agent incidents make the contradiction hard to ignore.
Read full story →Microsoft Says the First Fully Automated Ransomware Attack Already Happened
Microsoft's 2026 Digital Defense Report documents autonomous AI attacks as a present-tense fact, from a fully automated extortion campaign to a 32-step network takeover in the lab. The deeper warning is a speed gap, flaws weaponised in under 24 hours against patch cycles that take weeks, that AI cannot close on the defender's side.
Read full story →The Model Is the Easy Part: Enterprise AI's Real Bottleneck Reveals Itself
A Wall Street hiring surge, two cautious incumbents and a new benchmark all land on the same conclusion: the frontier model has become a commodity, and the money and difficulty now sit in data, governance and integration. "Agent orchestration" is the hottest skill on Wall Street; a smarter model still cannot repair a missing link between two databases.
Read full story →