Cybercriminal groups are promoting a new subscription-based AI service called Luciferus, which claims to bypass the ethical guards mainstream models enforce. The service was spotted in late August 2026 by threat researchers monitoring underground forums.
What is Luciferus and how is it being sold?
Luciferus is advertised as an “uncensored” artificial intelligence assistant designed to accept requests that most AI platforms reject—particularly content linked to malware development. The offering was first spotted on August 24, 2026 on a forum known as Exploit. The service’s ad was posted under the alias “Optimus_Prime,” an account designated as being of coding interest. Analysts note it joined the forum April 18 and had contributed 21 messages by early September.
The messaging claims Luciferus runs on a proprietary model with around 120 billion parameters and operates without moral or technical constraints.
Pricing tiers and capabilities
Multiple subscription plans are on offer. One set (from the forum ad) includes three tiers named “Inquisitor” ($35/month), “Archdeviel” ($55/month), and “Prince of Darkness” ($75/month), along with a VIP-level “Individual Embodiment” tier. The VIP package includes a customized model trained on the customer’s own data, dedicated hardware resources, and control over response behaviors.
However, Luciferus’ public-facing site shows different pricing: a “Junior” plan at $22, “Middle” at $34.75, and “Pro” at $47.14. That site omits mention of the VIP option entirely. Researchers say they don’t have clarity on why these tiers differ between the forum and the website.
Evidence of malicious behavior
Researchers tested the system by asking the “Junior” tier for code to build a simple remote access trojan (RAT) in Python. The model provided a description in Russian of how such a trojan works, including networking and command execution functions, followed by source code. The analysts don’t know if the code is functional—it was not executed—but the willingness to comply is significant.
Luciferus is distinct from “jailbroken” versions of AI models that simply attempt to trick content filters. Its design suggests a more permanent removal of restrictions, possibly through custom or open-source model assumptions. Still, experts caution that claims of a fully proprietary model should be viewed skeptically, since true foundation models require large datasets, extensive computing power, and technical expertise.
Broader trends in illicit AI offerings
Luciferus is part of a wave of AI-driven tools tailored for cybercrime—following in the footsteps of services like WormGPT and FraudGPT. These have been marketed for phishing, email compromise, malicious scripting, and direct malware creation. Researchers have increasingly observed that criminal AI services are becoming more polished: tiered subscriptions, support channels, and ongoing updates are becoming standard.
There’s also growing concern about investor-like optimism in underground forums: some of these offerings might be overhyped, built on lightly modified open-source models or simply layered prompts rather than truly new technology. Skepticism around operational quality is spreading among criminal buyers.
Knowing that Luciferus was caught in a malware-request test and seems set up like a commercial platform is a warning sign: this isn’t just theorized misuse—it’s live and evolving.
Why this matters
The emergence of a model specifically designed to serve criminal intent without reform underscores an alarming shift in AI threat landscapes. Luciferus reflects how illicit operators are no longer trying to hijack mainstream tools but instead are building parallel, unfettered products.
This trend raises urgent questions for AI policy, platform safety, and law enforcement: if unrestricted models like Luciferus gain traction, the ability to regulate or limit malicious usage becomes far more difficult. Monitoring underground markets won’t be enough; defending against models that are purpose-built for wrongdoing may demand new technical, legal, and ethical frameworks.
