Criminal AI Storefronts Use Shared Backend for Resilience
The use of a centralized routing layer allows illicit services to bypass safety filters and remain operational.
Updated on Oct. 22, 2026 in Artificial Intelligence
Live Poll
Should companies that provide AI infrastructure be held responsible for the content users generate through them?
Security firm ThreatDown reports that three criminal AI marketplaces—Kriminal, DadGPT, and WormGPT—are using the OpenRouter platform as a shared backend. These storefronts act as wrappers that route user prompts to various large language models through OpenRouter's API.
Why it matters
The reliance on a shared routing layer allows illicit operators to maintain service continuity even if individual storefronts are taken down. By blending criminal traffic with standard commercial usage, these actors significantly complicate efforts by developers to detect and block malicious requests.
The architecture involves 3 identified storefronts using OpenRouter to access models including Grok, Mistral Large, and Llama 3.3. This centralized endpoint allows operators to switch between models dynamically to bypass individual safety filters.
The players
ThreatDown
A cybersecurity research firm focused on identifying threats within emerging AI infrastructure and illicit digital marketplaces.
OpenRouter
A service that provides a unified API endpoint for accessing various third-party large language models from different providers.
The details
Storefront operators utilize OpenRouter as a gateway to rent access to established commercial large language models (LLMs — systems trained on vast datasets to generate human-like text) rather than developing their own infrastructure. The shared routing layer enables operators to maintain service uptime; if one model refuses a prompt due to built-in safety constraints, the system automatically redirects to an alternative model that may lack equivalent restrictions.
Timeline
October 21, 2026: ThreatDown released the findings regarding criminal AI infrastructure usage.
The Tech Race
This architecture signals a tactical evolution in illicit services that mirrors the transition from self-hosted sites to resilient, proxy-based distribution networks. It moves the conflict from attacking static servers to identifying malicious usage patterns within global API traffic.
Developers and AI providers should be aware that their models may be unknowingly servicing criminal requests through third-party aggregators. This shift means security teams must move beyond monitoring specific domains and start analyzing prompt traffic patterns for malicious intent.
The takeaway
The use of shared API gateways is accelerating, making the infrastructure behind criminal AI marketplaces increasingly difficult to isolate. Watch for shifts in model provider policies regarding how their APIs can be resold or routed through third-party platforms.
Further reading
Learn more about the latest challenges in Artificial Intelligence safety and governance.
Source note: This article includes information reported by IT Brief Australia.
Live Poll
Should companies that provide AI infrastructure be held responsible for the content users generate through them?
