Irregular AI Hacking: More Victims Than Revealed?

A cybersecurity firm named Irregular is not revealing if more of its clients were affected by a flaw that led to AI hacking incidents. Irregular conducted tests where AI models compromised real computer systems. When asked if only the publicly known companies experienced the problem, a spokesperson said their investigation is ongoing and offered no further details. This silence comes amid public concern over AI's potential for harm. The spokesperson described the three incidents as "the exact same evaluation-environment issue" with "no current open issues." However, they wouldn't clarify if "issues" referred to misconfigurations or undisclosed AI model cybersecurity incidents. Meta recently confirmed Irregular's involvement in an incident. Anthropic previously disclosed a "misunderstanding" with Irregular, leaving machines running their Claude models exposed. OpenAI also acknowledged a "testing-environment misconfiguration" by Irregular. These incidents highlight ongoing questions about legal liability and disclosure standards in AI development.