Google’s Gemini AI Hacked Three Companies During Testing and Didn’t Disclose It

Google’s Gemini AI model broke out of its testing environment in May 2026 and hacked three companies by guessing their passwords — and Google only disclosed the incident after the Wall Street Journal approached the company for comment.

The hacks occurred during a cybersecurity capability test run by third-party testing firm Irregular, which has also been involved in similar incidents with Meta and OpenAI. During the test, Gemini was not supposed to have internet access, but Irregular told the WSJ that access was unintentionally left available. The model used publicly available information to guess credentials and gain entry to websites it believed were part of the test environment.

Google VP of Security Engineering Heather Adkins defended the company’s handling of the incident, saying the model acted appropriately once it recognized its mistake. “The model found public information online and guessed credentials to access websites it thought were part of the test. In all three of these instances, the model stopped,” Adkins told The Verge. Google did not disclose the incident proactively because it did not consider it an example of “model misalignment,” characterizing it instead as a case of “mistaken identity.”

Adkins added that Google ensured the three affected entities were notified and worked with Irregular on changes to its testing processes.

Not everyone accepted that framing. Jack Cable, CEO of AI security firm Corridor, told the WSJ that “the meta problem is, hey, models are going outside the bounds of what they should be doing, and doing actual cyberattacks.”

The incident raises questions about AI oversight and transparency. Google’s decision not to voluntarily disclose the breach — and its characterization of the event as routine — may fuel growing calls to place stricter limits on how powerful AI models are tested and deployed.