A newly observed technique has revealed how attackers are using X’s AI assistant, Grok, to distribute malware by slipping past the platform’s ad restrictions.
The method, referred to as “Grokking,” hides malicious links in the metadata of promoted video posts. When prompted with targeted replies, Grok unintentionally exposes these links to a wide audience, effectively becoming an accomplice in the attack.
The exploit takes advantage of unscanned metadata fields and the built-in trust associated with Grok’s official account. Harmful links bypass platform rules while appearing to come from a legitimate, system-trusted source.
Why It Matters: This case shows how AI can be manipulated into amplifying malicious activity. It exposes a gap in moderation systems; trusted AI responses are not held to the same scrutiny as other content. The combination of automation, metadata use, and platform trust creates a scalable way to deliver malware without triggering defenses.
- Unscanned Metadata Fields: Attackers embed malicious links in the “From:” metadata field of promoted video ads, a field currently ignored by X’s ad-scanning systems. Since the visible ad contains no obvious links, it easily passes review, allowing attackers to pay for placement without raising suspicion.
- Prompting Grok for Disclosure: Once the ad is live, attackers ask seemingly harmless questions like “Where is this video from?” Grok interprets the metadata as part of its response, surfacing the hidden link publicly. The AI unintentionally becomes the messenger for the attacker. The result is a clickable malicious link posted publicly by a system-trusted account.
- AI-Generated Responses Increase Reach: Because Grok is viewed as a trusted system actor, its responses carry more weight than a typical user post. The links are more likely to be seen, shared, and indexed by search engines, increasing their exposure far beyond the original ad.
- Coordinated Campaigns: Researchers found hundreds of accounts repeating this tactic in rotation. Each account posts the same style of promoted videos until it is suspended, ensuring the campaign remains live at scale despite takedowns.
- Final Destinations: The links often redirect to fake CAPTCHA challenges, credential-harvesting forms, or malware downloads. Many are funneled through traffic distribution systems and shady ad networks, dynamically changing content depending on a user’s location, device, or profile.
Threat actors abuse X’s Grok AI to spread malicious links – Bleeping Computer
Trusted insights for technology leaders
Our readers are CIOs, CTOs, and senior IT executives who rely on The National CIO Review for smart, curated takes on the trends shaping the enterprise, from GenAI to cybersecurity and beyond.
Subscribe to our 4x a week newsletter to keep up with the insights that matter.
