North Korea is already using Gemini to hack computers around the world

A North Korean hacker group known as UNC2970 has begun using Google’s Gemini artificial intelligence model to carry out reconnaissance and cyberattack activities. This behavior has been documented in a report from the Google Threat Intelligence Group (GTIG), which highlights a concerning intersection between professional research and malicious activities in the field of cybersecurity.

AI is fine, man

According to reports, UNC2970 has focused on companies in the cybersecurity and defense sector, resorting to open-source intelligence (OSINT) to profile high-value targets and prepare specific phishing campaigns. The group uses Gemini to map technical roles and their salaries, creating profiles that facilitate the design of targeted attacks.

Researchers have identified new types of malware that use Gemini, such as HONESTCUE, a framework that allows attackers to generate additional functionality code, and COINBAIT, a phishing kit designed to masquerade as a cryptocurrency exchange, aimed at collecting credentials. HONESTCUE, for its part, operates by using the Gemini API to receive source code in C#, which is executed in memory, leaving few traces on the target system.

Additionally, ClickFix campaigns have been detected that use AI-generated instructions to solve common computer problems, but ultimately distribute malware designed to steal information. Experts warn that assuming keeping the model weights private is enough for protection is a serious mistake; gathering information through queries can result in the replication of the model’s original behavior.

Google has taken steps to disrupt these attacks and has highlighted the growing risk posed by hacker groups that use tools like Gemini to accelerate the phases of the cyber attack cycle. As technology advances, so do the tactics of those looking to exploit it.