After OpenAI, Anthropic Says Claude Also ‘Gained Unauthorised Access’ To Real World Systems
· inc42.com
After reviewing 1.41 lakh cybersecurity evaluation runs, Anthropic disclosed that three Claude models accessed real-world production systems because of a misconfigured third-party testing environment, unlike OpenAI's sandbox escape caused by a zero-day exploit
Anthropic suspended the evaluations, tightened safeguards and warned other AI labs to review testing environments, with the incidents adding to growing concerns...