IncidentAI-enabled attacks
Custom GPTs used to push ClickFix malware to at least 40 victims
Huntress found two malicious Custom GPT instances titled "Plus 5.6" that answered prompts with a link to a ClickFix page, tricking users into pasting a PowerShell command. The chain delivered a remote access trojan through a signed application with DNS-over-HTTPS command and control. Victims found the fake GPTs through sponsored search results. OpenAI removed the first on September 25; a second appeared two days later. An earlier Huntress case used ordinary ChatGPT and Grok conversation links in search results to deliver a macOS stealer.
Replay the attack step by step in the Learning Center
How it happened
- Attackers published a Custom GPT and bought search ads so people looking for ChatGPT Plus found it first.
- The GPT answered any prompt with a link to a ClickFix page that asked the user to run a command to 'fix' something.
- The command installed a remote access trojan hidden behind a signed application.
Impact
At least 40 people ran attacker commands because an AI assistant told them to.
The fix
OpenAI removed the GPTs; organizations should block script execution from clipboard prompts and treat AI-delivered instructions as untrusted.
Lessons
- People trust commands that come from an assistant; attackers now use that trust as the delivery mechanism.
- Marketplace listings and sponsored results need the same scrutiny as email links.