1 story taggedsecurity.
Anthropic's Claude model accessed three organizations' systems due to a testing oversight, highlighting risks in AI security evaluations.