AIToday
Large Language ModelsAI Safety & AlignmentZenn AI/MLPublished: Nov 20, 2026, 04:00 JST

ClickFix abuses real ChatGPT and claude.ai chats to plant commands

Send on LINE
ClickFix abuses real ChatGPT and claude.ai chats to plant commands

3 Key Points

  1. What happened

    Huntress, Malwarebytes, Trend Micro and Microsoft reported ClickFix cases where fake error or CAPTCHA screens, including attacker-made ChatGPT, Grok and Claude shared chats, push users to run commands themselves.

  2. Why it matters

    Because the malicious instructions sit on genuine domains and the user launches the code personally, advice like "don't open suspicious files" and automatic defenses may not catch it, so verifying where a command came from matters more than trusting the domain.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The reports summarized here trace a shift in where malicious commands appear. Microsoft's security blog, cited as published on August 21, 2025, says ClickFix was first observed between March and June 2024, and Japan's National Police Agency also covered it in an October 2025 cyber police newsletter. More recently, Huntress, Malwarebytes and Trend Micro have documented cases where the lure is a conversation or screen hosted on a real AI service. In the macOS cases, search results or search ads pointed users to attacker-made ChatGPT, Grok or claude.ai shared chats that presented terminal commands as a "safe cleanup" or as setup steps for "Claude Code on Mac", and Trend Micro says the shared conversation was disabled after its report. A September 2026 Huntress case on Windows went further: a malicious Custom GPT only redirected users to another domain, where a fake CAPTCHA page pushed a PowerShell command, so no command appeared inside the AI chat at all. The common thread the author draws is that the AI did not spontaneously produce the malicious command; rather, attacker-made content was published on genuine domains, which suggests users should treat the origin of a procedure, not the brand of the domain, as the thing to verify. A legitimate official install step can itself take the form of curl ... | bash, so the deciding question is whether the steps were found in official documentation opened directly or arrived via an ad or shared conversation.

FAQ
What exactly is ClickFix?
It is a technique that shows fake error or confirmation screens, such as a "I am not a robot" check, and gets the user to copy and run a command themselves.
When did Microsoft first observe this technique?
Microsoft's security blog says it first observed the technique between March and June 2024.
What does the Malwarebytes and Trend Micro report say about the claude.ai case?
Search ads led to a shared chat on claude.ai that posed as setup instructions for "Claude Code on Mac" and had users run commands; Trend Micro says it could steal SSH keys and crypto-asset files, and the shared chat was disabled after the report.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleAI人事: AI for hiring, evaluation and labor tasks