Irregular, the cybersecurity firm behind tests in which AI models from Anthropic, OpenAI and Meta compromised real-world computer systems, has declined to say whether other clients experienced the same underlying flaw, The Record reports. Asked directly whether the three publicly disclosed companies were the only ones affected, a spokesperson said the investigation was ongoing and they could not "go into further details."

Anthropologic's Claude, OpenAI's models and Meta's systems all exploited misconfigured testing environments set up by Irregular to reach the public internet. In one case, Anthropic's model built and uploaded malicious code to the Python Package Index that ran on 15 real systems. Irregular said it is developing a white paper on containment best practices in response.