Demand for artificial intelligence-based tools has surged on underground forums and specialized platforms. These tools help carry out phishing attacks, impersonate other people, write malicious code, and bypass the safety restrictions of popular models. In just a few months, the number of listings offering such services has grown from a few dozen to more than a thousand per month.

According to Halcyon’s Ransomware Research Center, underground platforms featured fewer than 50 listings for AI tools per month at the end of 2025. By February, that number had exceeded 1,400. Researchers analyzed nearly 4,000 posts across 77 Telegram channels, 20 dark web forums, and five specialized underground marketplaces from October through May, Axios reports.

From hacked accounts to specialized AI

The range of offerings on the underground market has expanded significantly. While cybercriminals were still relying mostly on separate, fragmented tools in 2023, a full-fledged commercial ecosystem is now forming around AI.

One segment consists of hacked or stolen access credentials for popular AI services. On underground platforms, for example, ChatGPT Plus accounts were being offered for as little as 10 cents, along with ready-made prompts for bypassing model safety restrictions. Some of these instructions were sold for about 32 cents.

Another class of tools is designed for identity fraud and phishing. AI makes it faster to create messages and other materials needed to impersonate a specific person or organization.

Models trained specifically on malicious data are also being sold separately. As an example, researchers cite WormGPT, a system geared toward generating phishing emails and malicious code.

There are also more specialized solutions. In particular, AI tools and infrastructure have appeared on the underground market to help automate specific stages of fraud campaigns. One service identified by researchers is capable of calling up to 120 people simultaneously.

Cybercriminals quickly adopted the SaaS model

According to the researchers, what matters is not only the growing number of tools, but also how they are being sold. Cybercriminals have effectively adopted the commercial models of the legitimate tech business: subscriptions, ready-made services, and online storefronts.

Some developers first advertise and test their tools on traditional dark web forums, then move them to Telegram shops, where they are easier to manage and sell to a larger number of customers.

Cynthia Kaiser, head of the Ransomware Research Center and former deputy assistant director of the FBI’s cybercrime division, notes that criminals have quickly mastered not only the AI models themselves, but also the ways to commercialize them.

As a result, specialized capabilities that once required certain technical skills are becoming cheap, ready-made services. This lowers the barrier to entry into cybercrime: a person no longer necessarily has to develop a complex tool independently if the needed function is sold like a standard service.

AI is not yet hacking systems fully on its own

At the same time, the current underground market does not mean that universal AI hackers have emerged that can independently carry out an attack from start to finish.

Halcyon researchers did not find offerings that would allow a complex cyberattack to be fully organized and executed autonomously. Even in the case of AI tools for malware, they are mostly aimed at individual tasks: writing code, preparing infrastructure, or automating specific actions.

According to Kaiser, AI today primarily makes existing attacks cheaper and more effective. It helps criminals speed up individual stages of already familiar schemes rather than turning one inexperienced user into a fully autonomous hacker.

This, the researchers believe, is the main change. The danger is not the emergence of a universal hacking machine, but the gradual commercialization of many specialized AI tools. Each of them handles a small part of the task, but together they create an increasingly accessible infrastructure for cybercrime.