AI · 3h ago
Sophos found an Exploit forum ad on August 24 for “Luciferus,” a subscription uncensored large language model (LLM) sold to cybercriminals as an alternative to jailbroken ChatGPT or Claude. The poster, “Optimus_Prime,” claimed it could answer without moral or ethical restrictions and offered several paid tiers, while Sophos said it could not verify the ad’s performance or privacy claims.
The key difference is how it is packaged: instead of waiting for a public chatbot to be tricked past its guardrails, the service is presented as built without those safeguards from the start. Sophos also assessed with low confidence that it may be based on Qwen, an open-source model from Alibaba, and said a test prompt produced malware code and an explanation of how it worked.
If the service is genuine, the lasting issue is not one forum post but a repeatable supply of harmful code that can outlive jailbreak fixes. For teams using AI assistants for code or content, the relevant exposure is the broader abuse surface around model access, prompt records, and output filtering, because the same convenience can be offered to criminals as a service.
1 source covering this story
An uncensored AI service called Luciferus is being sold as a subscription on a hacking forum, capable of generating malware, Sophos found.
Part of the PlainSec briefing for 2026-09-15