Anthropic called the incidents involving Claude a "failure of operational security," saying they occurred due to errors in a third-party evaluation environment. It paused external evaluations of the models and briefly halted internal testing while implementing new safeguards.
firstpost.com
