Fake ChatGPT and Claude Apps Hijack PCs to Mine Monero

How poisoned searches deliver phony AI clients

Criminals manipulate search engine results to place fake GitHub project pages ahead of legitimate software portals.

Curiosity around generative AI updates remains a goldmine for digital con artists. According to a telemetry report from Check Point Research, an active cybercrime campaign has spent the past two months targeting people searching for standalone desktop clients.

Rather than building convincing GitHub repositories filled with artificial stars and forks, the attackers leaned entirely on search engine optimization (SEO) poisoning. By distributing vast arrays of cheap blog posts and links across the web, the group convinced Google's crawlers that their repositories were authoritative. As a result, users searching for tools like Claude Opus or fantasy builds like "GPT-5.6 Sol" and "GPT-6 Astra" saw malicious GitHub pages right at the top of their search results.

The lure exploits a simple habit: people prefer standalone desktop software over browser tabs. When victims run the packaged installer, no AI model greets them. Instead, the payload immediately drops a loader dubbed Yoba.

Inside the Yoba loader and XMRig payload

Yoba goes to considerable lengths to stay hidden on Windows machines. It integrates directly into startup routines to guarantee persistence after a reboot while disguising its background process under normal Windows operational names.

To throw automated security sandboxes off track, the malware can intentionally idle for two full weeks before executing its next stage. It also interrogates the host operating system's keyboard configuration. If the system detects keyboard layouts from Russia, Belarus, Kazakhstan, or Uzbekistan, the program quietly exits without deploying a payload—a signature trait of Eastern European cybercrime groups seeking to avoid domestic law enforcement.

Once a system passes those clearance checks, Yoba pulls down its primary payload: an XMRig cryptominer. Unlike noisy ransomware, XMRig runs silently in the background, consuming processing cycles to mine the privacy-centric cryptocurrency Monero. The victim rarely notices anything beyond unprompted fan noise, sluggish system performance, and a higher electric bill.

A broader pattern of AI impersonation

Cryptojacking might seem mild compared to credential harvest schemes or corporate extortion, but loaders like Yoba represent an unpredictable risk. Because Yoba maintains an open command-and-control conduit, the attackers can replace the miner with credential stealers, infostealers, or corporate ransomware at any moment.

The security sector has repeatedly sounded alarms over attackers exploiting organizational interest in automation. Similar operational pivots were highlighted when AI agents breached Anthropic systems through external software integrations, as well as recent advisories where institutions like Bank of America warned users against giving AI agents direct credentials. Whenever a software category sees massive public demand, malicious actors reliably flood the ecosystem with fake installers.

What it means for you

If you frequently test new productivity tools, stay clear of third-party distribution portals for major language models. OpenAI and Anthropic provide their desktop clients exclusively through their primary official domains and verified platform app stores (such as the Microsoft Store and macOS App Store).

Be skeptical of desktop clients advertising unannounced model tiers like GPT-6. If you notice persistent high CPU utilization, unusually loud fans while idling, or suspect a past download, check Task Manager for unfamiliar high-resource tasks and scan your device with an up-to-date endpoint security tool.

As long as major lab releases generate massive headlines, malicious search campaigns will try to intercept curious users. Treating unverified desktop wrappers with deep suspicion remains your best defense against having your PC turned into a criminal mining node.