Luciferus: Uncensored AI of 120 billion parameters is sold in hacking forum

Sophos researchers identified Luciferus, an uncensored AI service sold on a hacking forum as a direct substitute for ChatGPT and Claude jailbreaks. The product presents itself as a proprietary model of 120 billion parameters capable of responding without the ethical restrictions imposed by large laboratories — and is offered precisely to those who do not want to waste time bypassing the locks of commercial assistants. The case was detailed by Help Net Security.

What is Luciferus, the uncensored AI sold on a forum?

Luciferus is neither a jailbreak nor a malicious prompt: it is a complete AI service, with its own model, announced in a clandestine space where offensive tools already circulate naturally. According to Sophos, the commercial appeal is to eliminate the weakest stage of the abusive use of AI — the attempt to break the protections of closed models, which usually fail, be corrected by providers and leave traces of account and billing.

The offer fits into a market that had already been forming: AI-based malicious tools are sold openly in clandestine forums, with showcases, support and periodic updates.

Why does 120 billion parameters matter to uncensored AI?

Parameters are the gross measurement of the capacity of a model. Encryptions in the hundreds of billions house place systems in the range of large open models capable of running on high-cost infrastructure or on rented servers. In practice, the number indicates that Luciferus is not an amateur fine-tuning experiment: it requires real computational power, which suggests operation with funding and structure behind it.

There is no disclosure of information on price, volume of subscribers, time of operation or identity of those responsible — relevant gaps to measure scale.

What's Luciferus' connection to Qwen from Alibaba?

Sophos points out that Luciferus may have been built on Qwen, a family of Alibaba models. The hypothesis is consistent with a known pattern: Chinese open models are often chosen as the basis for "downgraded" versions, because their weights are available for adjustment and redistribution — something that closed models, such as ChatGPT and Claude, technically do not allow.

This connects the case to a broader movement. Free and private models that "never say no" already conquer users, as shown by our coverage of IA without restrictions outside major providers.

Why does uncensored AI matter to Brazilian devs and companies?

On the defensive side, the existence of unrestricted models packaged as a product lowers the entry cost of those who want to generate phishing, malware or abuse content in scale — including in Portuguese, if the system supports the language. On the side of compliance, it reinforces a point that legal teams already face: blocking domains of known laboratories does not cover the use of generating AI within a company.

For those who build products, the message is due diligence of supply chain. If a model supplier does not explain the origin of the weights, there is no way to ensure that it does not derive from an unrestricted basis or from data obtained irregularly.

What the uncensored AI announcement does not solve

Luciferus doesn't eliminate the risk to the operating party. Selling computational capacity and AI traffic leaves footprints of infrastructure, payments and hosting — points that researchers usually follow. Nor is it clear whether the service actually delivers unrestricted answers or just a marketing layer about a common open model, nor if Sophos was able to test it directly. Without this data, what exists is an ad, not a proven ability.

Frequently Asked Questions about Luciferus and Uncensored AI

What is Luciferus?

It is an uncensored AI service identified by Sophos and sold on a hacking forum as an alternative to ChatGPT and Claude jailbreaks, using a proprietary model of 120 billion parameters.

Why use an unrestricted model instead of a jailbreak?

Because it eliminates the step of breaking the protections of closed models, which often fails, is corrected by providers and leaves traces of account and billing.

Is Luciferus based on Alibaba's Qwen?

Sophos points out that the information was not publicly confirmed. Chinese open models are common bases for unfiltered versions for allowing adjustment and redistribution.