A newly surfaced criminal AI service named MessiahGPT is being marketed on BreachForums as an unrestricted offensive model capable of generating ransomware, phishing kits, stealers, crypters, rootkits, and social-engineering content on demand.
The Trellix Advanced Research Center has reported that this service operates through the domain messiahgpt[.]de and promotes an associated Telegram community, marking a shift from informal “jailbreak” bots to a more structured cybercrime-as-a-service platform.
Unlike mainstream generative AI products that utilize safety measures such as Reinforcement Learning from Human Feedback and policy-based model controls, MessiahGPT is advertised as having no ethical or legal restrictions.
Its operators claim that the model was trained from scratch on unrestricted manuals, dark-web archives, leaked documents, and unfiltered internet data.
While these claims, including information about its technical architecture, cannot be independently verified, the public promotion, accessible platform, and marketing on criminal forums indicate a clear intent.
The service is attributed to Dabial Leaks, a cybercrime community known for database sharing and disruption activities, as well as cooperation with hacktivist-aligned groups, as reported by Undercode Testing.
MessiahGPT Unrestricted AI Model
MessiahGPT is said to use a 128-expert Mixture-of-Experts architecture, with 16 experts active per token. This architecture is designed to route requests to specialized model components, potentially enabling lower operating costs compared to a uniformly active model.
However, there is no independent evidence supporting the advertised number of experts, model size, or training process.
The commercial design of MessiahGPT raises greater concerns than its technical claims. It reportedly offers 50 free queries without requiring user registration, with paid access starting at around $8 per month.
Payments are accepted only in cryptocurrency without know-your-customer (KYC) verification. This low-friction model could allow inexperienced users to experiment with malicious prompts and scale attacks without requiring advanced malware development skills or established access to criminal marketplaces.
The threat posed by this service lies not necessarily in the sophistication of every generated payload. Instead, the platform may lead to an increase in the volume and variety of phishing lures, credential-harvesting pages, scripts, and malware prototypes.
AI-assisted phishing can be tailored to specific industries, job roles, current events, and local language patterns, making older detection filters, previously reliant on poor grammar, repetitive templates, or known malicious text, less effective.
| Feature | Reported details | Defensive relevance |
|---|---|---|
| Access model | 50 free queries; crypto-paid subscriptions | Lowers barriers for opportunistic actors |
| Claimed architecture | 128-expert MoE; 16 active experts per token | Not independently verified |
| Promoted output | Phishing kits, ransomware, stealers, crypters, rootkits | Expect rapid code and lure variation |
| Platform tiers | JinnatGPT, ParaohaGPT, MessiahGPT 2.0 | Suggests segmentation from basic to advanced users |
| Distribution | BreachForums, website, Telegram | Enables discovery, support, and customer acquisition |
Security teams should not consider simply blocking the domain as a complete mitigation strategy. While blocking messiahgpt[.]DNS security controls, secure web gateways, firewall egress policies, and threat-intelligence feeds can reduce direct access from managed endpoints, but attackers may still use proxies, alternative infrastructure, copied outputs, or independently hosted phishing content.
Therefore, organizations should monitor DNS logs for newly observed suspicious domains, flag unusual outbound connections, and investigate endpoint activity such as mass file modifications, abnormal archive creation, credential dumping, and suspicious scripting engine execution.
YARA rules can provide supplementary coverage, but they should target concrete malware behavior and file characteristics rather than attempting to identify “AI-generated” code.
Effective detection logic may focus on suspicious import combinations, embedded encoded content, encryption APIs, persistence mechanisms, process injection behavior, and known command-and-control patterns.
Behavioral endpoint detection and response remains more resilient than static signatures, especially when attackers can easily regenerate payload variants.
MessiahGPT operates within a broader underground market that includes services like DarkGPT, another openly promoted “uncensored” AI offering.
The key takeaway is that defenders can no longer assume an attacker’s capability is directly linked to coding skill. Organizations should prioritize phishing-resistant multi-factor authentication (MFA), email authentication, sandboxing, least privilege access, immutable backups, DNS visibility, egress controls, and rapid behavioral detection to mitigate the impact of AI-assisted attacks.
Stop new phishing & malware before they compromise your business. Integrate live intel from 15K SOCs around the world
