Threat Actors Exploit ChatGPT and Luma AI Popularity to Poison Search Results and Deliver Malicious Payloads

Threat actors are leveraging the surging popularity of artificial intelligence platforms such as ChatGPT and Luma AI to orchestrate sophisticated malware distribution campaigns.

According to recent research by Zscaler ThreatLabz, attackers are employing Black Hat SEO tactics to poison search engine results for trending AI-related keywords, thereby luring unsuspecting users to malicious websites that serve as launchpads for malware payloads.

Technical Attack Chain

The campaigns begin by crafting AI-themed websites using platforms like WordPress, meticulously optimized to rank highly for AI-centric search queries such as “Luma AI blog” or “ChatGPT download.”

When a user clicks on one of these top-ranking results, they are redirected to a deceptive landing page. Here, malicious JavaScript is triggered, initiating a multi-stage redirection chain.

The script fingerprints the browser collecting details like version, window resolution, cookies, and user agent and encrypts this data with a randomly generated XOR key before transmitting it to an attacker-controlled domain (e.g., gettrunkhomuto[.]info).

The attacker’s server decrypts and validates the data, then issues a 302 redirect to an intermediate site.

According to Zscaler Report, this site further checks the public IP and, based on predefined criteria, redirects the user to the final payload delivery page.

The malicious payloads observed in this campaign include Vidar Stealer, Lumma Stealer, and Legion Loader, all of which are distributed as password-protected ZIP archives containing large installer files.

This packaging is designed to evade sandbox analysis and detection systems that flag smaller, more typical malware executables.

Payload Analysis

Vidar and Lumma Stealer are delivered via 800MB NSIS installers, which embed files with misleading extensions (e.g., .docm) that are actually components of the malware.

Upon execution, these installers reconstruct an AutoIT loader and obfuscated scripts, which then deploy the stealer payload.

The installers also perform antivirus process checks (targeting products like Quick Heal, Webroot, Sophos, BitDefender, Avast, AVG, Norton, and ESET) and attempt to terminate any detected security software before proceeding.

Legion Loader follows a similar multi-stage process, using a password-protected ZIP archive and an MSI installer.

During installation, a custom DLL (DataUploader.dll) collects system information and communicates with a command-and-control (C2) server.

The loader then retrieves a dynamic password from the C2, extracts additional payloads via 7zip, and uses DLL sideloading and process hollowing to inject malicious code into legitimate processes such as explorer.exe.

In observed cases, the final payload included a browser extension for cryptocurrency theft.

The malicious JavaScript is often hosted on trusted CDNs like AWS CloudFront, making detection and takedown more challenging.

The scripts are also equipped with adblocker detection mechanisms; if an adblocker is found, the redirection chain is aborted to minimize exposure to security researchers and automated analysis environments.

Configuration details and redirection domains are stored in custom Base64-encoded strings within the JavaScript, further obfuscating the campaign’s infrastructure.

The attackers’ use of large installer files, dynamic password retrieval, and sideloading techniques demonstrates a high level of sophistication aimed at bypassing both automated and manual analysis.

This campaign highlights the growing trend of leveraging popular AI-related search terms as an entry vector for malware distribution.

The combination of Black Hat SEO, advanced redirection logic, and multi-stage payload delivery makes these attacks particularly insidious.

Users searching for AI tools should exercise caution and rely on official sources, while security teams must remain vigilant for evolving SEO poisoning tactics and layered malware delivery chains.

Indicators of Compromise (IOCs)

Find this Story Interesting! Follow us on LinkedIn and X to Get More Instant updates