Kriminal AI service bypasses guardrails by renting legitimate AI models

As detailed in Silicon Angle, a new report from ThreatDown, the business security arm of Malwarebytes, reveals that Kriminal, a popular tool in the criminal artificial intelligence market, is essentially a storefront that rents and repackages legitimate AI services.

Kriminal operates on the clearnet, offering various subscription tiers and pay-per-message options. Instead of developing its own AI, Kriminal utilizes services like SpaceXAI's Grok, Anthropic's Claude, Mistral Large, and Meta's Llama 3.3, routing them through OpenRouter and accessing live web search via Tavily. The service bypasses AI safety guardrails by employing a system prompt that instructs the underlying models to ignore previous instructions and safety policies. This approach allows Kriminal to offer unrestricted code generation, exploit development, and social engineering tools.

The operation's structure, relying on multiple legitimate vendors for different functions, makes it difficult to shut down, as each vendor only sees its own component of the service. This commoditization of AI tools for criminal purposes highlights a growing trend, where sophisticated AI capabilities are becoming increasingly accessible to malicious actors, posing a significant challenge for cybersecurity defenders.

Source: Silicon Angle