Mistral AI's and xAI's commercial large language models Mixtral and Grok, respectively, have been leveraged to create various jailbroken artificial intelligence tools, or WormGPTs, peddled on cybercrime forums, reports CyberScoop.
Identified to be powered by Mixtral was the WormGPT variant promoted as "WormGPT / 'Hacking & UNCENSORED AI," which was touted as an AI-based LLM providing details on cyberattack processes, vulnerability detection, and defensive strategies, a Cato Networks report showed. On the other hand, xAI's Grok was found to support the "Uncensored Assistant" model on Telegram, which allowed the creation of phishing emails and credential-stealing payloads. "It appears to be a wrapper on top of Grok and uses the system prompt to define its character and instruct it to bypass Grok's guardrails to produce malicious content," wrote Cato Networks researcher Vitaly Simonovich. Both tools were reported to be under a subscription-based model, with prices ranging from $631 for an annual license to $5,740 for a private setup.
Identified to be powered by Mixtral was the WormGPT variant promoted as "WormGPT / 'Hacking & UNCENSORED AI," which was touted as an AI-based LLM providing details on cyberattack processes, vulnerability detection, and defensive strategies, a Cato Networks report showed. On the other hand, xAI's Grok was found to support the "Uncensored Assistant" model on Telegram, which allowed the creation of phishing emails and credential-stealing payloads. "It appears to be a wrapper on top of Grok and uses the system prompt to define its character and instruct it to bypass Grok's guardrails to produce malicious content," wrote Cato Networks researcher Vitaly Simonovich. Both tools were reported to be under a subscription-based model, with prices ranging from $631 for an annual license to $5,740 for a private setup.
