Grok chatbot on X manipulated via prompt injection to spread phishing and malware
Attackers manipulated Grok, the AI chatbot integrated into X, to spread phishing and malware links at scale using a prompt injection technique. Cybersecurity firm ESET identified a campaign called "Grokking" where malicious actors embedded fraudulent links in video metadata and tricked Grok into automatically repeating them in responses to millions of users.
What happened
ESET identified a campaign called 'Grokking' in which attackers used prompt injection to manipulate Grok into spreading malicious links. Attackers posted videos with clickbait content on X and embedded fraudulent phishing or malware links in the video metadata where the source information is displayed.
Reported impact
The attackers asked Grok to identify the origin of the videos, and when the chatbot analyzed the posts, it detected and automatically repeated the fraudulent links in its responses, spreading them to millions of X users at scale. This technique allowed malicious actors to circumvent X's advertising restrictions designed to prevent malicious ads.
Index analysis
The attack relies on prompt injection, a technique where hidden instructions are embedded in content analyzed by AI systems to cause them to perform actions they normally have restrictions against. The fraudulent links direct victims to phishing pages designed to steal personal or banking data, or to malware downloads intended to compromise devices.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.