OpenAI and Anthropic models went rogue during testing (again)
The UK's AI Security Institute says this is the first time it has "seen risks around autonomy and deception manifest this clearly."
Dan Thorp-Lancaster
Editor
About Dan
Daniel Thorp-Lancaster has been interested in tech since he got his hands on his first Windows 3.1 PC as a kid. He has been writing full-time about everything from Windows and Xbox to iPhones, Macs and Android for more than 10 years, starting after he graduated from Illinois State University. His work has appeared in publications like Wired, TechRadar, Windows Central and Android Central. After a brief stint venturing into writing about local services and home improvement for Yahoo Local, he joined the Yahoo Tech team in early 2026.
885
Articles on Yahoo
Latest articles
-
-
Apple is working on letting you copy and paste between your iPhone and Windows PCs
The feature is expected to come next fall, following a request from Microsoft in the EU.
-
Samsung pulls the plug on smart TV apps that shared users' internet
Samsung says it's banning smart TV apps that contain residential proxies that let strangers use the TV owner's internet connection.
-
Apple is reportedly struggling to keep MacBook Air in stock due to the global memory shortage
Apple is reportedly running into shortages of the MacBook Air as RAMageddon continues to make memory scarce.
-
Google Earth briefly let you generate AI images from satellite imagery — then Google pulled it
Google has added Nano Banana 2 image generation to Google Earth, so you can use it to "visualize history, create real estate plans, and more."
-
Anthropic says it discovered its AI broke into other companies' servers as well
Anthropic revealed this week that it has discovered its AI models broke into systems at three outside organizations during evaluations conducted as far back as April.
