Fake Google Gemini installer deploys Vidar Stealer, steals browser data

Key Takeaways

• Cybercriminals are exploiting the widespread interest in generative AI by distributing malware disguised as legitimate AI software installers.
• A recent campaign used a fake Google Gemini installer to deploy the Vidar information stealer, compromising browser credentials and other sensitive data.
• The attack chain leveraged a legitimate Google Colab page to lend credibility to the malicious download, redirecting users to a fraudulent “Windows Software Hub.”
• The incident, discovered by Darktrace, underscores the critical need for users to download software only from official sources and for organizations to implement robust behavioral detection.

Cybercriminals are capitalizing on the explosive demand for generative artificial intelligence tools by luring unsuspecting users into downloading malware. In a recent incident documented by security researchers, a malicious file masquerading as an installer for Google Gemini successfully deployed the Vidar information stealer, exposing sensitive browser data and saved passwords to attackers.

This attack did not originate from a traditional phishing email. Instead, it exploited common software search and download patterns, illustrating how threat actors can weaponize routine inquiries for AI applications into vectors for credential theft. This strategy mirrors other recent campaigns that have utilized deceptive AI installers and search engine manipulation to disseminate password-stealing malware, including incidents involving fake Claude AI pages that convincingly mislead users.

Security firm Darktrace first identified this malicious activity in July 2026 within a customer environment located in the Europe, Middle East, and Africa (EMEA) region. According to a report shared with Cyber Security News (CSN), Darktrace analysts linked the suspicious executable to the Vidar stealer after observing anomalous process behavior, unusual outbound network connections, and indications of browser credential exfiltration.

The incident highlights a growing trend in malware delivery, where attackers increasingly conceal harmful files behind trusted brand names, popular cloud services, and well-known AI platforms. This tactic relies on users lowering their guard when a download appears to be associated with a reputable platform.

Hackers Exploit Fake Google Gemini Installer

The infection sequence commenced when a user executed a file named Download_Google_Gemini_For_Windows.exe from their Downloads folder. Darktrace’s investigation revealed that searches for this filename led to a Google Colab page, which presented a download prompt, thereby lending a veneer of legitimacy to the operation.

Google Colab, a platform frequently utilized by developers and researchers, served as an effective staging ground for the malicious lure. The page then redirected visitors to a secondary website, designed to appear as a “Windows Software Hub,” where the fraudulent Gemini installer was made available for download.

While the investigation could not definitively recover HTTP or file-download telemetry to pinpoint the exact initial download source, encrypted sessions with Google Colab immediately preceding the executable’s launch strongly suggested direct user interaction with the hosted resource before proceeding to the secondary download site.

At the time Darktrace researchers examined the activity, the Google Colab page remained active, hosting a ZIP archive that contained the malicious binary. The archive also included a README file instructing users to run the program with administrator privileges and to add it to antivirus exception lists—a clear red flag indicating an attempt to bypass security defenses.

Upon execution, the program was identified as a newer variant of Vidar, compiled in Go. It established communications with infrastructure linked to Telegram and connected to suspicious external systems over port 443. This network behavior was crucial in allowing investigators to connect the fake installer to the broader credential-stealing campaign.

This campaign underscores that relying solely on trusted hosting platforms is insufficient to validate software downloads. Similar risks have emerged as attackers continue to exploit interest in AI through cloned repositories, deceptive setup guides, and malicious archives. A recent fake AI tool campaign further demonstrated how attackers can leverage legitimate-looking developer resources to distribute information stealers.

Browser Credentials and Incident Response

Vidar is a sophisticated information stealer designed to exfiltrate valuable data from compromised devices. In this particular incident, subsequent endpoint alerts confirmed activity consistent with the theft of browser credentials and other sensitive information. Stolen data, including saved passwords, browser-based session tokens, and related details, can grant attackers unauthorized access to email accounts, business services, and personal online profiles without requiring a password.

Shortly after the fake installer was executed, the affected device initiated contact with an external IP address. Through SSL certificate analysis, researchers identified a related IP address and a command-and-control domain associated with the malicious activity.

Darktrace successfully contained the incident by blocking communications with the suspicious infrastructure and quarantining the compromised endpoint. The automated response was triggered by anomalous device behavior that deviated from its normal activity baseline, rather than by the perceived legitimacy of the installer or its hosting platform.

For organizations, the key takeaway is to approach all AI software downloads with the same stringent caution applied to any other application. Users should only obtain Gemini-related tools from verified official sources, avoid downloads initiated from search results that lead to unfamiliar pages, and never follow instructions to disable or exclude antivirus protection.

Security teams should also prioritize investigating unusual browser credential access attempts, new executables launched from Downloads folders, and rare encrypted connections originating from recently installed programs. These controls are particularly vital as SEO poisoning campaigns targeting Gemini continue to leverage convincing search results and fraudulent installation guidance to ensnare users.

This campaign, by combining a fake Gemini installer, abuse of a trusted platform, and social engineering, delivered an established information stealer through a highly deceptive route. The broader threat extends beyond a single malware family; attackers are increasingly recognizing the high demand for AI tools as a reliable conduit for credential theft.

What You Should Do

• Verify Software Sources: Always download AI tools and other software directly from official vendor websites or trusted app stores. Avoid third-party download sites or links found in search results.
• Be Skeptical of Instructions: Never follow instructions that ask you to disable antivirus software, add programs to exclusion lists, or run executables with elevated privileges unless absolutely certain of the source and necessity.
• Implement Behavioral Detection: Organizations should deploy advanced endpoint detection and response (EDR) solutions capable of identifying anomalous process behavior, unusual outbound network connections, and suspicious credential access attempts, regardless of the perceived legitimacy of the initial download source.
• Monitor Downloads Folder Activity: Pay close attention to executables launched from the Downloads folder, especially if they are new or unexpected.
• Educate Users: Conduct regular cybersecurity awareness training to inform users about the risks of social engineering, fake software installers, and the importance of verifying download sources, particularly in the context of popular and emerging technologies like AI.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.