Hackers tried to steal Google's Gemini AI using this method
Hackers recently tried to make their own version of Google's Gemini AI by bombarding it with over 100,000 questions in different languages.
Their method, called knowledge distillation, let them learn how Gemini thinks—basically building a copycat model without stealing any code.
State-sponsored groups used Gemini during cyberattacks
Some state-sponsored groups took things further by using Gemini itself during cyberattacks.
China's APT31 used tools powered by Gemini to find system weaknesses in US organizations.
Iran's APT42 crafted phishing emails with help from the AI, and North Korea's UNC2970 analyzed defense sector employees using it.
Google is now blocking suspicious accounts
Google fought back by blocking suspicious accounts and tracking unusual behavior linked to model extraction.
Interestingly, most attempts came from private companies and researchers worldwide—not just government-backed hackers.
With AI copying getting easier and cheaper, protecting models like Gemini is now more important than ever.