A security gap in X’s Grok assistant is being used by attackers to deliver harmful links that would normally be blocked. The weakness lies in how the platform checks and displays links to users.
A researcher at Guardio Labs found that some advertisers publish videos with adult content to attract clicks but avoid placing a direct link in the ad itself. Instead, they hide the link in a small metadata field under the video card. That field doesn’t appear to be checked for threats.
When someone asks Grok where the video comes from, the assistant pulls information from the hidden field and responds with a live link. Because the reply comes from Grok, which functions as a system account, it carries more weight and people are more likely to trust it.
Tests of these links showed they often pass through obscure advertising networks before sending users to fraudulent sites. Some of the destinations pretend to be verification pages, others push downloads that carry malware, and some are designed to steal personal information.
The tactic, referred to as “Grokking,” gives malicious ads wider exposure. In certain cases, the reach extends to millions of impressions, effectively turning the assistant into a channel for spreading the same campaigns the platform was trying to restrict.
The researcher suggested that the problem could be reduced if X scanned all areas where links can be placed, blocked Grok from surfacing hidden links, and added checks so the assistant doesn’t share unverified addresses. The findings have been reported to X, but the company has not issued a public response on when the gap will be fixed.
Notes: This post was edited/created using GenAI tools.
Read next: England Moves to Block Energy Drink Sales to Under-16s
A researcher at Guardio Labs found that some advertisers publish videos with adult content to attract clicks but avoid placing a direct link in the ad itself. Instead, they hide the link in a small metadata field under the video card. That field doesn’t appear to be checked for threats.
When someone asks Grok where the video comes from, the assistant pulls information from the hidden field and responds with a live link. Because the reply comes from Grok, which functions as a system account, it carries more weight and people are more likely to trust it.
Tests of these links showed they often pass through obscure advertising networks before sending users to fraudulent sites. Some of the destinations pretend to be verification pages, others push downloads that carry malware, and some are designed to steal personal information.
The tactic, referred to as “Grokking,” gives malicious ads wider exposure. In certain cases, the reach extends to millions of impressions, effectively turning the assistant into a channel for spreading the same campaigns the platform was trying to restrict.
The researcher suggested that the problem could be reduced if X scanned all areas where links can be placed, blocked Grok from surfacing hidden links, and added checks so the assistant doesn’t share unverified addresses. The findings have been reported to X, but the company has not issued a public response on when the gap will be fixed.
Notes: This post was edited/created using GenAI tools.
Read next: England Moves to Block Energy Drink Sales to Under-16s
