@MaxForAI: Big congratulations to Gemini!! Their model has finally gone off the rails too The Wall Street Journal just revealed th…

Big congratulations to Gemini!! Their model has finally gone off the rails too 🤣

The Wall Street Journal just revealed that in May this year, Google’s Gemini, during a cybersecurity capabilities test, unexpectedly accessed the real internet and autonomously breached the systems of 3 real companies.

Specifically, the security firm in charge of the test, Irregular, set up a simulated environment for Gemini to attack a fictional company.

But the test environment wasn’t completely isolated from the internet, and the fictional company’s name happened to overlap with real-world companies.

So Gemini followed the task and went all the way out.

In one company’s system, it guessed the password itself; for the other two companies, it found usable credentials from public code repositories and logged in directly.

These operations weren’t step-by-step directed by researchers, but rather Gemini independently searching for targets, collecting credentials, attempting logins, and actually entering real-world systems.

However, according to statements from Google and Irregular, Gemini stopped further actions upon realizing it had entered real companies, causing no known damage; Google subsequently notified the affected businesses and U.S. federal agencies.

Congratulations to Gemini 🤣

The Wall Street Journal (@WSJ): Google’s Gemini accessed the internet and hacked other companies during a test, the first known example of the company’s AI autonomously committing such an act