# The AI That Named Itself After Crime and Listed on Google Anyway
There's a certain audacity to naming your product "Kriminal," marketing it with features called WRAITH and ARCHITECT, accepting only cryptocurrency, and then slapping a terms-of-service page on it that claims everything is for "research, creative, and educational purposes."
That's the situation the security industry finds itself in this week, after ThreatDown — Malwarebytes' enterprise arm — published research on what it calls "one of the newest and most popular tools in the criminal AI market." The platform isn't hiding on Tor. Type "Kriminal AI" into Google and it comes right up.
## Thirteen Dollars and No Guardrails
For $12.99 a month in cryptocurrency, Kriminal gives you access to a language model that has been deliberately stripped of the safety mechanisms that OpenAI, Anthropic, and Google spend considerable engineering effort building and patching. The pitch isn't subtle. The platform explicitly advertises what it won't do: refuse.
Higher subscription tiers unlock the features worth noting. WRAITH offers "social engineering & persona craft" — the AI equivalent of a con artist coaching service. ARCHITECT is described as an "offensive security & exploit expert." The platform also throws in OSINT scanning (names, addresses), uncensored image generation, cryptocurrency tracing, and what the site calls "guardrail-free conspiracy discussion," which is a strange thing to advertise alongside security tooling, but here we are.
The company claims 2,300 active users and 18,400-plus messages sent to the platform to date. Those numbers are self-reported and unverified, but even if they're inflated by half, that's a meaningful user base for a product that launched with a name that's a synonym for "criminal."
## The Gap Between Policy and Practice
This isn't the first "jailbroken" AI service to surface. WormGPT emerged in 2023. FraudGPT followed. DarkBard. EscapeGPT. The category has existed for a few years now, and the pattern is always the same: a model trained or fine-tuned to bypass content filtering, sold on underground forums, occasionally surfacing on the open web when operators get bold.
What's different about Kriminal is the SaaS packaging. Most of its predecessors looked like what they were — illicit products marketed through Telegram channels and dark web forums, with rough interfaces and frequent service disruptions. Kriminal presents itself like a startup. There's a pricing page. There are feature names with branding. There's a terms of service. It has the aesthetic of a legitimate software product, which makes the content underneath the wrapper more unsettling, not less.
The "research and educational purposes" clause is worth examining for what it reveals rather than what it says. No one running a legitimate red team or security research operation needs to pay $12.99 in crypto to a no-name AI service with a name like Kriminal. Red teamers have internal tooling, established vendors, and professional frameworks. The ToS language exists to create legal ambiguity, not to describe the actual customer base.
## What It's Actually Useful For
Social engineering is where this matters most practically. Crafting phishing emails, developing personas for pretexting calls, generating plausible cover stories for business email compromise — none of that requires a technically sophisticated model. It requires a fluent, confident, uncensored one. A model that will write a convincing email impersonating an HR department without flagging the request is more useful to a BEC operator than any exploit kit.
The OSINT scanning capability is a separate concern. Aggregating public information — names, addresses, relationships — is already possible through a dozen free and paid services. But wrapping it in a conversational AI that can synthesize that data and feed it directly into persona development is a meaningful workflow improvement for someone running an impersonation campaign.
Exploit development is probably the least immediately dangerous feature, despite the dramatic ARCHITECT branding. Writing functional exploits requires understanding that a language model alone doesn't provide, and the most capable uncensored models are still poor at producing working shellcode. That may change, but right now the social engineering use cases are the realistic threat.
## The Clear Web Problem
The more interesting question isn't whether Kriminal is harmful — it clearly provides capability that accelerates certain categories of attack. The question is what enforcement mechanisms exist for a service like this, and the honest answer is: not many.
Kriminal doesn't appear to be hosted in a jurisdiction with meaningful cybercrime enforcement. It takes cryptocurrency, so payment rails can't be easily cut off. Its infrastructure is presumably distributed enough that a single takedown request won't land it. And because its terms of service claim legitimate use, it occupies a legal gray zone that makes aggressive action complicated.
The clear web availability cuts both ways. It makes the service easier to find and use. It also makes it easier to research, monitor, and eventually pursue. Services that operate entirely underground are invisible to law enforcement and security researchers. Services that list on Google can be tracked.
---
## HackWire Analysis
Kriminal is being written about as a novelty, but it's a data point in a trend that's been accelerating since 2023: the commoditization of offensive AI capability below the technical skill floor traditionally required.
The earlier wave of criminal AI tools — WormGPT, FraudGPT — were rough products with limited reach, mostly circulating through Telegram channels and requiring some underground access to find. The packaging was underground. The friction was real. Kriminal represents a step change in accessibility. You find it on Google. You pay with a credit card equivalent. You have a support ticket system. The UX is consumer software.
This matters because it shifts the attacker profile. The operators running sophisticated BEC campaigns already had AI tools — they were using ChatGPT with jailbreak prompts, or had internal tooling. What services like Kriminal enable is the tier below: less technically sophisticated actors who benefit from a purpose-built, polished interface that removes the friction of working around safety filters.
For defenders, the immediate implication is that phishing and social engineering volume is going to continue rising, and the quality floor of those attempts is going to keep climbing. The badly-written phishing email is not disappearing, but it's going to represent a smaller share of what lands in inboxes. Security awareness training that relies on catching grammatical errors and awkward phrasing needs a significant update.
The harder structural problem is that there is no meaningful enforcement mechanism for services like this. Legal gray zone, crypto payments, opaque jurisdiction. The security industry's best response right now is detection: monitoring for WRAITH-style persona patterns in communications, treating OSINT-enriched phishing as the new baseline, and assuming that social engineering attempts are going to arrive better researched than they were two years ago.
— HackWire Editorial
---
## Related Coverage