The brief · 13 February 2026
Two days after GLM-5, research keeps pushing forward on every front: molecules seen as images, ChatGPT locking itself down, and AI making its way into the ballot box.
Prompt injection remains the black hole of AI agent security. OpenAI's answer is a mode that cuts ChatGPT's network access when you don't need it. In practice: if you use ChatGPT to browse or act on the web, switch it on for sensitive tasks.
practitioners › Lockdown Mode is opt-in and lives in settings, paired with Elevated Risk labels on capabilities deemed more exposed. Dated February 13, 2026.
OpenAI →Nothing mandatory here, just a voluntary framework for teaching AI skills at work and in schools. Still worth noting: this is exactly the kind of document that training providers end up copying wholesale for years. If you work in workforce development, it's worth reviewing what's inside before it hardens into the de facto standard.
practitioners › Five content domains and seven implementation principles, published February 13, 2026 by the Employment and Training Administration, non-binding guidance.
U.S. Department of Labor →Democratic candidates are making AI regulation a centerpiece of their 2026 campaign strategy, betting that public unease over AI's societal effects can be converted into electoral advantage against a Trump administration that opposes intervention. Whether this actually moves voters remains untested.
practitioners › The Trump administration is holding its hands-off line, with no new federal framework announced.
source non précisée →