Mandiant flags fake AI video generators laced with malware

As the internet fills up with clips from AI-video generators, hacking groups are seeding the online landscape with malware-laced programs and fake websites hoping to cash in on the trend.

Tracked by researchers at Mandiant and Google Cloud, the campaign is being carried out by a group identified as “UNC6032.” Since mid-2024, they have spread thousands of advertisements, fake websites and social media posts promising victims access to popular prompt-to-video AI generation tools like Luma AI, Canva Dream Lab and Kling AI.

Those promises lead to phishing pages and malware, with the group deploying infostealers and backdoors on victim devices. Compromised parties saw their login credentials, cookies, credit card data and in some cases Facebook information stolen, and the scheme appears to be impacting a wide range of industries and geographic areas.

“Mandiant Threat Defense has identified thousands of UNC6032-linked ads that have collectively reached millions of users across various social media platforms like Facebook and LinkedIn,” wrote researchers Diana Ion, Rommel Joven and Yash Gupta. “We suspect similar campaigns are active on other platforms as well, as cybercriminals consistently evolve tactics to evade detection and target multiple platforms to increase their chances of success.”

The emergence of highly realistic AI prompt-to-video generation tools over the past several months has generated curiosity, concerns and a significant amount of interest from the public. According to Google Trends, internet searches for AI video generation tools have surged over the past year, and especially since April.

The technology today is capable of creating startlingly lifelike people and scenes with virtually none of the glitching or visual cues that made previous AI-generated videos easier to spot.

Cybersecurity company Morphisec, which published similar research earlier this month, noted how the proliferation of AI video generators over the past year has lowered the barrier for new entrants, giving even low-technical users the ability to create realistic fake media. The rush to jump on this latest trend, from users who may not be highly technical or familiar with AI tools, represents a new opportunity for cybercriminals and hackers.

“What makes this campaign unique is its exploitation of AI as a social engineering lure — turning an emerging legitimate trend into an infection vector,” wrote Morphisec researcher Shmuel Uzan. “Unlike older malware campaigns disguised as pirated software or game cheats, this operation targets a newer, more trusting audience: creators and small businesses exploring AI for productivity.”

Mandiant researchers gave a shout-out to Meta, which was apparently aware of and investigating UNC6032’s campaign before being notified by Mandiant, and contributed to the research. Using Meta’s ad library, which has enhanced ad targeting information for European users due to regulations, Mandiant’s team found more than 30 different websites that were cited in thousands of fake ads, mostly on Facebook through attacker-created pages or hacked accounts.
Nearly all the websites advertised free or high-quality AI-video generation capabilities.

“Once the user provides a prompt to generate a video, regardless of the input, the website will serve one of the static payloads hosted on the same (or related) infrastructure,” the researchers wrote.

Google Cloud has said UNC6032 has a “nexus” to Vietnam. Mandiant and Google Cloud use the term “UNC” to denote unique clusters of hacking activity for which there is only limited available information and telemetry. 

That means UNC6032 may be an offshoot of a previously tracked threat group using different tactics, techniques and procedures or a completely new hacking group, and while the activity has a “nexus” to Vietnam, that does not necessarily imply a state-based connection.