[Digital Today intern reporter Seung-a Yoo] A warning has been issued that users looking for how to install Claude Code on a Mac could have their cryptocurrency wallets stolen after following a fake guide page.

On Aug. 9 local time, IT outlet TechRadar reported that the attack centred on combining Google search ads with Anthropic's public shared-link feature to make it look like a legitimate guide.

After searching on Google, a victim clicked a paid ad shown at the top of results and was taken to a step-by-step installation guide page within claude.ai, a shared link for Anthropic's Claude. The page appeared as if it were a document shared by "Apple Support" and prompted the user to open Terminal and paste a one-line command.

Security firm Huntress said an analysis of the attack confirmed a six-stage macOS malware chain called MacSync. It included an information-stealing malware, a remote access trojan and a signed helper program designed to gain specific system privileges. In the final stage, it was designed to swap out the victim's existing cryptocurrency wallet application with a trojanised version to extract the recovery phrase.

Researchers said they did not directly obtain the malware left on disk because the victim took the device offline. Instead, they re-downloaded and analysed each stage from the delivery server used by the attacker and assessed the attack as "a sophisticated campaign with very little that needed to be forged to appear legitimate".

The case is drawing attention because it gained trust without a fake domain or certificate warnings. Anthropic allows anyone to publish a conversation as a public shared link, and the attacker exploited that feature. The lure page was served over HTTPS from the claude.ai address with Anthropic's security certificate applied. The structure made it hard to spot suspicious signs from the address bar alone.

The attacker also set the display name to Apple Support. As a result, a safety banner at the top of the page showed the phrase that the user was "viewing a copy of a conversation between Claude and Apple Support". The platform repeatedly displayed the name chosen by the attacker as if it were fact, adding credibility. The body of the guide was also written like an official vendor document, explaining that the installation process does not affect personal files and does not make system-level changes without approval.

The issue is also tied to the structure of public shared links. Anthropic's shared links are collected by search engines like ordinary public webpages. There have been past cases in which Claude conversations believed to be private appeared in Google and Bing search results. Both cases in which users accidentally expose their own conversations and cases in which attackers plant fake installation guides stem from the same characteristic: public shared addresses are "public, indexed and served under Anthropic's certificate".

Similar methods are not new. Huntress said it had previously confirmed cases in which attackers tainted ChatGPT and Grok conversations to spread AMOS, distributed a separate remote access trojan using fake Claude desktop ads, and posted fake installers for other artificial intelligence tools on GitHub.

The case shows that the process of searching for installation information for AI tools can itself become a target. The method has changed in some parts, but the broad outline remains. Users should recheck the source and intent before trusting links or commands found online, even if they appear at the top of Google ads. In particular, installation guides that prompt users to paste commands directly into Terminal require additional verification even if they use a real service address.