1. News
  2. Alerts
  3. Fake AI · Security
AlertFake AI · Security2 min read

Fake Claude and ChatGPT downloads are spreading password-stealing software

Researchers logged 1,258 fake AI tool scams in six months. Many ask you to paste a command that quietly installs software to steal logins, code and accounts.

By the Dashline deskSource: HackerNoon
Fake Claude and ChatGPT downloads are spreading password-stealing software
Photo: Pauli Rautakorpi, CC BY 2.0
Act today
What happenedResearchers found fake websites posing as AI tools that trick people into running a command or signing in, which installs software that steals data.
Who is affectedDevelopers, marketing teams and any employee who searches for or installs AI tools such as Claude, ChatGPT, GitHub Copilot, Gemini or Ollama.
What to do
  1. 1Download AI tools only from the vendor's official website, not from a search result or ad.
  2. 2Never paste an install command into PowerShell or Terminal unless you trust the page it came from.
  3. 3Be wary of shared AI chats and sign-in pop-ups that ask for your work login.
  4. 4If you ran a suspicious command, tell your IT team right away and change your passwords and keys.

Fake websites posing as Claude, ChatGPT, GitHub Copilot, Gemini and other AI tools are tricking people into installing software that steals data. Security firm ANY.RUN tied 1,258 public test analyses to these scams between April 15 and October 5, 2026. The US ranked second after India as a source of those submissions.

What happened

The fake pages look like real download or sign-in pages. Many tell you to copy an install command and paste it into PowerShell or Terminal, the text windows that control a computer. The researchers found that at least 45% of the analyses came from campaigns that use this trick. Others use fake installers or fake sign-in windows.

Claude and ChatGPT are the names scammers use most. Claude made up 601 of the analyses, or 48%. ChatGPT and Codex made up 418, or 33%. Together that is 81%.

Why it is hard to spot

You are the one who runs the attack. Because you paste the command yourself, some security tools may not stop it. The lures also sit on well-known sites, including GitHub, Google Sites, Cloudflare Pages, and even claude.ai and chatgpt.com. They reach people through:

  • Search results and Google Ads
  • Shared AI chats
  • Fake download pages
  • Email invitations

The researchers say the outcomes include password-stealing software, remote-control software, stolen browser sessions and phishing of business accounts. On a work computer, that can expose source code, cloud and AI keys, ad accounts and billing.

Who it affects

Developers and marketing teams are the most exposed. Developers often hold code and cloud keys. Marketing teams often hold ad and billing accounts. Any employee who searches for and installs AI software on their own is also at risk. The US made up 69 of 422 country records, or 16%.

People who only get AI tools from the vendor's official site, and never paste commands from a page they do not trust, avoid the main trick described here.

What to do

The alert box above lists the steps. The short version: go to the official site yourself, do not trust ads or search results, and tell IT right away if you ran something suspicious.

What is not known

The 1,258 figure counts analyses submitted to a public testing service. It is not a count of victims, and the material does not say how many people were infected or how much was lost. It also does not name the specific stolen-data programs involved. What is clear is that the campaign ran every month from April to September, with more than 100 analyses each month.

The takeaway

Get AI tools only from the vendor's official site, and never paste an install command from a page you do not fully trust.

More articles

You should really read these.

Next step

Let’s talk