Anthropic says its models went rogue and hacked 3 companies during testing

Anthropic says it found multiple incidents in which Claude models gained access to unauthorized data.

Bloomberg/Getty Images

Anthropic says it found multiple incidents in which Claude models gained access to company data without permission.In a blog post on Thursday, Anthropic said it proactively conducted a large-scale review of its cybersecurity systems following an incident last week in which OpenAI models accessed parts of Hugging Face's live systems.The AI lab, which has filed to go public this year, said that it reviewed more than 141,000 AI tests and found three cases where Claude models got online during testing and accessed the live systems of three organizations without authorization."In all cases, Anthropic's evaluation prompt specified to Claude that its environment was a simulation and that it had no internet access," the blog read. "Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available," it added, referring to Irregular, an AI security startup.The lab said three different Claude models were involved in these cases, which started in April: Opus 4.7, Mythos 5, and an internal research test mode.It added that it has reached out to the three affected organizations to remediate. Two of the organizations it has reached were not aware of the accidental hack.Anthropic told Business Insider it did not have a comment beyond the blog post.