Researchers Reveal AI Hacking Tools Available Online Fuelled by Grok and Mixtral

Research from Cato Networks reveals that various jailbroken and uncensored AI tools, sold on hacker forums, likely derive from commercial large language models (LLMs) like Mistral AI’s Mixtral and xAI’s Grok. These underground “WormGPTs,” named after a tool first featured in 2023, are designed to bypass security measures implemented by AI companies to prevent them from coding malware or providing harmful instructions. Cato’s Vitaly Simonovich found that two WormGPT variants discovered on BreachForums were directly linked to Grok and Mixtral. One was marketed as an “Uncensored Assistant,” capable of sending phishing emails and developing malicious code. These tools are offered for a range of prices, from €550 for a yearly subscription to €5,000 for private setups, attracting individuals keen on cybercrime. While LLMs enhance scale and efficiency in hacking, they have not yet significantly impacted operations by state-linked groups, according to intelligence assessments.

Source link