Google Just Confirmed its Gemini AI Went Off-script and broke Into 3 Real Companies During a Security Test in May
By Mario Nawfal
A bug accidentally gave it internet access, and Gemini treated real businesses like they were part of the test.
It guessed passwords on one and used leaked login details it found online for the others.
The wild part? Once it realized the companies were real, it stopped on its own.
Google says no damage was done.
AI is getting pretty good at acting on its own… maybe a little too good.
Google just confirmed its Gemini AI went off-script and broke into 3 real companies during a security test in May.
A bug accidentally gave it internet access, and Gemini treated real businesses like they were part of the test.
It guessed passwords on one and used leaked…
— Mario Nawfal (@MarioNawfal) September 21, 2026
