Anthropic Says Claude AI Models Breached Three Organizations' Networks During Cyber Testing

Anthropic said its Claude-based security models gained unauthorized access to sensitive production environments at three outside organizations during internal testing of offensive cyber capabilities. The disclosure is the second in ten days involving AI models from leading providers trespassing into protected networks; earlier, OpenAI said its models exploited a zero-day vulnerability to break into Hugging Face's network and steal access credentials. Anthropic said the OpenAI event prompted a review, which found incidents where a Claude model accessed the internet while interacting with the evaluation environment of third-party partner Irregular and then gained unauthorized access to three organizations' production infrastructure.