Fake ChatGPT and Claude Downloads Are Hiding Crypto Miners

Key points

• A campaign tracked as Yoba is distributing malware disguised as desktop apps for ChatGPT and Claude, using fake GitHub projects with names like GPT-5.6 Sol, GPT-6 Astra and Claude Opus.
• The downloads install a loader that quietly pulls down a cryptocurrency miner, which uses the victim's computer to generate digital coins for the attackers.
• Neither OpenAI nor Anthropic ships a general-purpose desktop app under the names used by the fake projects.
• The installers carry valid-looking code-signing certificates, which helps them slip past basic Windows warnings.
• Attackers are increasingly baiting victims with AI tool branding rather than cracked software.

Criminals are hiding cryptocurrency miners inside fake desktop versions of ChatGPT and Claude, and pushing them through GitHub pages built to look like real projects.

The campaign, which researchers are calling Yoba, was detailed this week by The Hacker News. It targets people hunting for a tidy desktop app for the two best-known AI chatbots. There isn't one, at least not under the names being used, and that's the gap the attackers are stepping into.

What exactly is being installed?

A loader, which is a small program whose only job is to fetch and run more malware, followed by a cryptocurrency miner. The miner runs on the victim's processor to generate digital coins for the attackers, slowing the machine and hiking the electricity bill while the user notices little beyond a hot laptop.

The fake projects carry names designed to sound like plausible next-generation releases: GPT-5.6 Sol, GPT-6 Astra, Claude Opus. Each links to a Windows installer. Once run, it drops the loader, which then reaches out to an attacker-controlled server and pulls down the miner.

Why do the downloads look trustworthy?

Because they are signed. Windows checks installers for a code-signing certificate, a digital stamp asserting that a named company stands behind the file. The Yoba installers carry valid-looking signatures, so the usual red warning most people rely on never appears.

That's the part worth paying attention to. A signed installer from a GitHub repo with a convincing name defeats the two checks most ordinary users actually perform.

Is there a real desktop app I should be using?

OpenAI publishes a ChatGPT app for macOS and Windows through its own site at chatgpt.com, and Anthropic's Claude is primarily a web product at claude.ai with a limited desktop client of its own. Anything calling itself GPT-6, GPT-5.6 or a numbered Claude build on GitHub is not from either company.

If you have installed something matching those names recently, uninstall it, run a full scan with Microsoft Defender or your preferred security tool, and check for sluggishness or fan noise when the machine is idle, both classic signs of a hidden miner.

How this fits the wider pattern

For most of the past decade, this kind of bait sat on torrent trackers wrapped around cracked Adobe or pirated games. AI branding is now the lure of choice: that is what people are searching for, and the pool of users willing to trust an unfamiliar installer is bigger than the security industry would like. We covered the earlier shift to this playbook on 17 September, when attackers abused Claude Artifacts and shared ChatGPT links to reach users who trusted the branding.

Watch the signing. If the attackers behind Yoba are obtaining valid certificates repeatedly, somebody in the issuance chain is either being tricked or not looking closely enough, and that problem extends well beyond this one campaign.