Cybercriminals are weaponising the global obsession with artificial intelligence tools to distribute malware through fake social media ads, according to a new report from Mandiant Threat Defense.
The report, released today, details an ongoing campaign run by a Vietnam-linked threat group known as UNC6032. Since at least mid-2024, the group has been creating fraudulent websites that imitate popular AI video generator platforms, including Luma AI, Canva Dream Lab, and Kling AI, to distribute malware designed to steal user data.
Victims are typically lured through malicious ads on platforms like Facebook and LinkedIn, which redirect them to spoofed websites that appear to offer AI-powered services such as text-to-video or image-to-video generation. Instead of generating content, the websites deploy malware—specifically, a modular toolkit known as STARKVEIL—which installs a series of information stealers and backdoors.
Once infected, users’ credentials, credit card numbers, cookies, and other sensitive information are quietly exfiltrated, often via the Telegram messaging platform. Mandiant researchers believe the stolen data is being sold on underground marketplaces.
“This campaign is a striking example of how threat actors evolve with the digital landscape,” said Yash Gupta, Senior Manager at Mandiant Threat Defense. “They’ve hijacked the excitement around AI and paired it with social engineering tactics, creating legitimate-looking ads and websites that are anything but.”
Scale and Sophistication
The scale of the operation is significant. Mandiant identified thousands of malicious ads, primarily on Facebook, with many promoted using either attacker-created business pages or compromised user accounts. A sample of more than 120 ads targeting users in the European Union revealed a combined estimated reach of more than 2.3 million people.
While Facebook was the primary platform, similar—albeit smaller—campaigns were also observed on LinkedIn. One fake domain, klingxai[.]com, was registered in September 2024 and promoted via LinkedIn ads within 24 hours of registration. Some of these ads garnered between 50,000 and 250,000 impressions, with most of the traffic originating from the United States, Europe, and Australia.
The group behind the campaign rotates domains frequently—often launching new sites and ad campaigns daily—to avoid detection by security systems and evade platform moderation.
Growing Threat Amid AI Boom
Mandiant’s findings come amid a broader surge in AI adoption, where user interest has created fertile ground for scams and malicious exploitation. According to the company’s latest M-Trends 2025 report, stolen login credentials are the second most common method for initial cyber intrusion, further underscoring the risks posed by campaigns like that of UNC6032.
“These fake AI tools don’t just target consumers—they can compromise enterprise credentials, giving attackers access to sensitive systems and corporate networks,” Gupta warned.
Industry Collaboration
Mandiant worked with both Meta and LinkedIn to investigate and disrupt the campaign. Meta, according to the report, had already begun removing many of the malicious ads and domains prior to Mandiant’s involvement—a proactive response that Mandiant praised. However, with new ads appearing daily, ongoing collaboration will be critical to containing the threat.
Similar findings were recently published by cybersecurity firm Morphisec, suggesting a broader consensus on the campaign’s scope and impact.
Staying Safe
Security experts urge users to be cautious when engaging with ads for new or trending tools, particularly AI applications. Recommendations include:
-
Visiting official websites directly rather than clicking on ads
-
Verifying URLs before downloading any content
-
Using updated antivirus software
-
Reporting suspicious ads to the platform immediately
“AI is transforming the digital landscape,” Gupta said. “But that same innovation is now being mirrored by attackers. It’s more important than ever for users to stay vigilant.”
