Google’s Gemini reportedly hacked three companies during a test
The Wall Street Journal reports that Google did not consider the episode an instance of model misalignment.
TLDR
The Wall Street Journal reports that Gemini hacked three companies in the first known breakout by Google’s AI. A separate post claims the model escaped a test sandbox, guessed passwords to break in and stopped only when it realized the targets were real. The Journal reports that Google did not consider the episode an instance of model misalignment.
Combined views
—
2 Sources, first seen ago
