Fake AI Downloads Distributed Via Google Ads Target Mac Users

A new cybersecurity threat has emerged, leveraging Google advertisements and Microsoft's Bing search engine to distribute malicious software disguised as legitimate AI assistant downloads. Security researchers at Push Security discovered a campaign that tricks Mac users into downloading a fake version of Claude, an AI assistant developed by Anthropic.

The Deceptive Search Engine Pathway

The attack begins when users search on Google for "Claude Mac." A sponsored search result, appearing to direct users to Bing, is displayed. However, clicking this ad initiates a chain of redirects. The user's click is processed by Google's advertising system, then Bing's click-tracking service forwards the browser to a compromised South American retailer's WordPress website. This compromised site ultimately directs the user to a malicious domain, claude-desk-code[.]com, which mimics the official Claude download page.

This multi-stage redirection, which researchers have dubbed "Adception," is designed to obscure the malicious origin by routing traffic through seemingly legitimate services. The fake download page is presented convincingly, making it difficult for users to suspect the deceptive journey they have taken.

Cloaking and Deception Techniques

Adding another layer of sophistication, the fake Claude website employs cloaking techniques. It checks the visitor's referral source and browser information. Users who arrive directly at the malicious site might see a harmless 404 error, making it harder for security scanners to detect the threat. This ensures that only users arriving via the carefully constructed redirect chain encounter the fake installer.

Hidden Malicious Commands

The danger escalates when a user attempts to download the fake Claude application. The website displays an installation command that appears to be from Anthropic. However, when the user clicks the "Copy" button, a different, malicious command is secretly placed on the clipboard. This command uses Base64 encoding to hide an attacker-controlled website.

Upon execution in the Mac's Terminal application, the command displays a reassuring message about downloading Claude from its official site. Simultaneously, it decodes the hidden address, downloads a remote script, and executes it directly on the user's Mac. This technique, a variation of "ClickFix" scams, tricks users into unknowingly running potentially harmful code.

While the exact final payload of the executed script has not been publicly detailed by researchers, running an untrusted command in Terminal can grant malicious code extensive permissions. This could potentially lead to data theft, file modification, or the installation of further malware. Push Security has identified other websites utilizing similar malicious toolkits.

Industry Responses and User Protection

Google has confirmed that it has suspended the advertiser accounts involved in this campaign, stating that distributing malware through ads is a severe policy violation. The company indicated that attackers use cloaking to evade detection systems and that it continuously updates its defenses.

Microsoft also responded, with a spokesperson confirming that Microsoft Defender protections identify and block the malicious indicators associated with the reported campaign.

To protect against such threats, cybersecurity experts advise users to:

• Download AI software directly from the official developer's website, such as Anthropic for Claude.
• Be highly cautious of any website that instructs you to copy and paste commands into your Mac's Terminal.
• Verify the exact command before execution by pasting it into a plain text editor to check for hidden or suspicious elements.
• Keep operating systems and antivirus software updated, and utilize real-time protection.
• Be skeptical of sponsored search results and research advertisers through platforms like Google's Ads Transparency Center.

Users who believe they may have executed the fake Claude command are advised to disconnect their Mac from the internet, run a security scan with reputable antivirus software, change critical passwords, enable two-factor authentication, and check for suspicious applications or system settings. If sensitive information is involved, professional assistance may be necessary.