Google says no damage was done. The more useful question for builders is how a test ended up touching the real internet.
Google Gemini AI Hacked Three Real Companies During a Test (Source: https://www.kaspersky.com/resource-center/preemptive-safety/gemini-hacked-3-copmanies )
On September 18, Google confirmed that its Gemini model accessed systems at three external companies during a test, and the Wall Street Journal first reported it. Google said the access happened in May and came from guessing login information or using credentials found in a public repository.
If you build or evaluate AI agents, the interesting part is not the headline. It is how a test came to touch the real internet, and what you can do so yours never does.
What was disclosed
The incidents took place during cybersecurity testing conducted by Irregular, an AI-focused security company.…