November 21, 2026

Fake Claude Desktop Download Hosted on Anthropic’s Own Website Infected Dozens of Businesses

MacBook Pro turned-on

A malicious download page built to look like the official Claude Desktop app was hosted directly on Anthropic’s claude.ai website earlier this week, tricking people into installing data-stealing malware, according to new research from cybersecurity firm Huntress.

The page was viewed more than 7,100 times before it was taken down, and at least 29 organisations ended up infected in the space of two days, Huntress said.

How the scam worked

People searching online for “Claude Desktop app”, the desktop version of Anthropic’s popular AI chatbot, were shown a sponsored search advert that pointed to claude.ai, Anthropic’s real website. That part of the link was genuine. But rather than leading to Anthropic’s actual download page, it pointed to a public “artifact”, a type of user-generated page that people can publish and share on Claude’s platform, that had been disguised to look exactly like the official app download screen.

Anyone who clicked “Download” was quietly redirected to a series of look-alike websites before being served a file named ClaudeDesktop.exe. Rather than installing the real app, this file installed a well-known password- and data-stealing programme called SectopRAT, designed to grab saved logins, credit card details, and files stored on the victim’s computer.

Huntress reported the fake page to Anthropic, which removed it on 22 July.

Sophisticated behind the scenes

While the trick itself was fairly simple: a fake download button on an otherwise trustworthy-looking site, researchers found the malware behind it was unusually advanced. It was built to detect when it was being examined by security researchers, checking a computer’s graphics hardware for signs it was running inside a virtual testing environment before deciding whether to activate.

The programme also hid its instructions inside cryptocurrency transactions on the Ethereum blockchain, a technique that makes it far harder for investigators or internet providers to shut down its command centre, since new instructions can simply be posted as a fresh transaction rather than hosted on a server that can be taken offline.

Huntress said it used Anthropic’s own Claude AI model to help unpick some of the malware’s more complex encryption, calling it a notable example of AI being used to help investigate an attack that was, in turn, distributed by piggybacking on the same company’s platform.

Investigators also traced the person behind the campaign to a similar scam earlier this year involving a fake version of Docker Desktop, another popular developer tool, as well as infrastructure previously flagged by Microsoft in a global takedown of malware distribution networks.

What people should do

Huntress is urging people to be cautious with sponsored search results even when they appear to point to a legitimate, well-known website, and to download software directly from a company’s official homepage rather than via search adverts. As AI tools continue to grow in popularity, researchers say scammers are increasingly using searches for AI apps as bait to distribute malware.

Anthropic has not issued a separate public statement on the incident at the time of writing. Huntress’ full technical report, including a list of technical indicators associated with the campaign, is available here: https://www.huntress.com/blog/fakeagent-claude-desktop-malvertising-ends-in-dotnet-rat