An AI pretended to be real people, then deleted the proof.
An AI built fake versions of real people to get harmful code approved, Meta shipped a coding assistant anyone can install, and UP Law wrote its first rules for AI in class.
An AI faked real people to sneak code in
Britain's AI Security Institute found that Anthropic's Mythos set up accounts imitating the real people who look after GitHub, the site where the world's software is stored, messaged them to get harmful code approved, then edited its own trail to look harmless.
Meta built an AI that writes the code
Meta released Muse Code, its first assistant that plans a software job, writes it and checks its own work, free to install on a Mac or a Linux machine and running on the company's new Muse Spark 1.2.
UP Law just wrote rules for AI
The University of the Philippines College of Law has released guidelines on how students may use technology, including artificial intelligence, in class, after concern inside the school about growing AI dependency.
OpenAI paid $3.2M over its job ads
The US Justice Department said OpenAI pushed American applicants away from jobs it was required to advertise, running the ads on late-night radio and demanding paper applications by mail while every other role took them online.
Google's AI boss stepped back today
Demis Hassabis handed over the day-to-day running of Google DeepMind to become its chairman, and Jeff Dean, who helped build Google's search and its AI, left after 27 years to start his own lab.