Photo: Koshiro K / Shutterstock.com
Cybercriminals took over several Facebook pages and redesigned them to impersonate popular generative AI models, including Midjourney, Sora, ChatGPT and Dall-E; then ran ads to spread password-stealing malware to millions of people following these pages.
These malicious Facebook pages are designed to lure unsuspecting people into downloading the non-existent official desktop version of these popular AI models and, in turn, deliver malicious payloads that can harvest sensitive information, including login credentials, autocomplete data, credit card and crypto wallet information.
The malvertising campaigns served ads that contained links to infected executable files serving Rilide, Vidar, IceRat, and Nova Stealers. These campaigns actively targeted people in Belgium, France, Germany, Italy, Netherlands, Poland, Romania, Sweden, and Spain, among other countries.
Once a Facebook page was successfully taken over, cybercriminals would change its descriptions, cover and profile photo to make it look like an official one, run by one of the above-mentioned AI generators. The page’s legitimacy was boosted by posting news, AI-generated photos, and ads with valid descriptions of these AI services, as well as links offering free trials or desktop versions of the AI.
Once someone clicks on this link, the malware springs into action, downloading their personal information and stealing credentials.
One such Facebook page, discovered by researchers at Bitdefender, impersonated Midjourney and had over 1.2 million followers.
The ad campaigns run on this Midjourney page — one of the dozen pages impersonating Midjourney — reached approximately half a million male Facebook users aged between 25 and 55. While most of these malicious payloads were uploaded to Dropbox or Google Drive, the Midjourney campaigns mimicked the official AI’s landing page and redirected to a GoFile link. The malicious payload available for download here was Rilide v4, which impersonated the Google Translate extension, hiding the malware and siphoning data in the background.
The page, hijacked in June 2023, was active for over a year and was only shut down on March 8, 2024. But since then, the cybercriminals behind these campaigns have increasingly set up more fraudulent pages to deliver malicious payloads, one of which has over 600,000 followers.
The data stolen from the victims of such campaigns is often compiled and sold on dark web marketplaces and forums. Threat actors can even breach these people’s accounts and scam or defraud them through them.
These AI models’ Facebook ad campaigns peddling malware in the garb of non-existent apps and services isn’t an isolated incident; on the contrary, such campaigns have been running amok. Earlier this month, malicious Google Ads campaigns were discovered delivering malware. Google has also rolled out the Device Bound Credentials feature to prevent cookie theft, as with the malware spread through the Midjourney campaign.
In the News: China can target Indian, US, and South Korean elections: Research
