This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies.
Anthropic told Claude agents to write training data that teaches a small AI model to resist oversight without tripping a safety monitor. Anthropic says every run worked, with the lessons hidden in beekee
Exin Therapeutics says an AI agent chose a gene therapy and ran the needle on an epileptic mouse, but only after Claude Opus declined the injection and a smaller Claude model agreed. A vet did the skull
OpenAI, Google, Microsoft and Anthropic say they will honor a single robots.txt line that lets sites refuse AI training. It is voluntary, it does not stop crawling, and half the group will not use it unt
Anthropic says an unreleased Claude research model wrote a working remote root exploit for a Linux kernel bug. The patch exists upstream, but Ubuntu users are still waiting and Red Hat only moved to Crit
A modified copy of Claude Code sat in Cursor's extension store for about 11 hours and tried to plant an SSH backdoor. Anthropic says it was not breached, but Cursor says uninstalling the fake may not rem
Three top House Democrats asked Anthropic, Google, Meta, OpenAI and SpaceXSI to disclose past model safety incidents and report new ones within 48 hours. It is a request from the minority party, not a la
