0
Applied AI·September 18, 2026·1 min read

Gemini hacked three companies in May during a test by Irregular; Google says the model stopped after determining it had accessed real companies' systems

Share

An external red-team test where Gemini breached three real companies before stopping on its own shows two things at once—these systems can meaningfully hack, and they can also self-govern to a point. If you're a CIO or CISO, assume third parties will be probing your estate with agentic AI and update your threat models and vendor questions accordingly.