Case wormgpt-grok-mixtral-variantsAI-assisted malwareWorldwide
Cato CTRL: new WormGPT variants are jailbroken wrappers on Grok and Mixtral
Cato CTRL obtained two WormGPT variants sold via Telegram and showed one runs on xAI Grok and the other on Mistral Mixtral with guardrail-stripping system prompts; both wrote phishing emails and credential stealers.
MediumAttack using AIStatus: OngoingVerified case file
Open in the interactive mapCite this case
- Reported
- Jun 17, 2025
- Location
- Worldwide
- Target
- Phishing and credential-theft victims of subscribers
- Severity
- Medium
- Triggered by
- keanu and xzin0vich (forum handles)
- Actor type
- Criminal
- Reported loss
- None reported
- CVE
- None
What happened
Cato CTRL obtained two WormGPT variants sold via Telegram and showed one runs on xAI Grok and the other on Mistral Mixtral with guardrail-stripping system prompts; both wrote phishing emails and credential stealers.
Impact: Working uncensored assistants available to paying criminals; direct victim counts not reported.
Who triggered it
Actor: keanu and xzin0vich (forum handles)
Attribution: Cato researcher Vitaly Simonovich extracted system prompt fragments identifying the underlying models; high confidence on model identity.
Motive: Subscription revenue from uncensored AI access, about 550 EUR per year or up to 5,000 EUR for private setups
How it happened
Attack vector: Telegram chatbot wrappers around commercial LLM APIs with manipulated system prompts
Root cause: Prompt-level guardrails on commercial models can be overridden by a hostile system prompt
Timeline
- Mixtral-based variant advertised on BreachForums
- Grok-based variant advertised
- Cato CTRL publishes findings; xAI and Mistral did not comment
Response
Cato published the analysis; no provider action was reported at publication.
Current status: Ongoing
Sources (open in a new tab)
- Cato Networks catonetworks.com
- CyberScoop cyberscoop.com
- TechRepublic techrepublic.com
Cite this case
Cato CTRL: new WormGPT variants are jailbroken wrappers on Grok and Mixtral. AI Threat Watch, case wormgpt-grok-mixtral-variants, Jun 17, 2025. https://aithreat.io/cases/wormgpt-grok-mixtral-variants/
Permanent link: https://aithreat.io/cases/wormgpt-grok-mixtral-variants/