A report from Halcyon Ransomware Research Center shows widespread advertising of AI-based tools on underground cybercrime markets, with monthly listings rising from fewer than 50 in late 2025 to more than 1,400 by February.
What the research examined
- Halcyon analyzed nearly 4,000 posts across 77 Telegram channels, 20 dark web forums and five specialised underground markets between October and May.
- The study found offers such as access to ChatGPT Plus accounts for as little as $0.10 and jailbreak prompts—designed to remove safety guardrails from mainstream models—advertised for $0.32.
- The report also includes examples of other offerings at higher price points (for instance around $10) for prompts or services that remove model safeguards, illustrating a range of price points for different products.
Types of AI tools being sold
According to the report, sellers are offering roughly four categories of products:
- Jailbroken and stolen AI services, including prompts that let users bypass safety restrictions on mainstream models.
- Identity fraud and phishing tools that simplify impersonation of real people.
- Access to models trained or configured for malicious purposes (the report cites WormGPT as an example, a model designed to draft phishing emails and malware).
- AI-augmented malware and infrastructure, such as systems capable of initiating large numbers of simultaneous calls as part of scam operations (the report mentions systems that can call up to about 120 people at once).
Threat level and impact
The research warns that the spread of these AI tools makes some formerly specialised tasks cheaper and easier, potentially lowering the barrier for criminals to perform parts of cyberattacks. Cynthia Kaiser, head of Halcyon’s Ransomware Research Center and former deputy director of FBI Cyber, told Axios that the price examples underscore how inexpensive it has become to conduct these activities — a worrying trend for organizations.
The report notes companies are already losing millions of dollars to attacks that are enabled or made easier by the kinds of tools observed on these markets.
Market dynamics
Halcyon finds that cybercriminals have commercialised both the sale and packaging of these tools, often imitating software-as-a-service subscription models and online storefronts. Some sellers have moved tools they tested on traditional dark web forums into easier-to-manage storefronts hosted on Telegram.
The researchers point out that whereas in 2023 only a handful of ad-hoc tools were available, the ecosystem has become more robust. Still, Kaiser emphasized that Halcyon found no evidence of criminals selling AI systems capable of conducting fully autonomous cyberattacks. Much of what is observed involves AI-assisted tasks — coding help, infrastructure automation, content generation — rather than AI creating and deploying malware on its own.
What to watch
- The rapid adoption of commercial sales practices by cybercriminals makes these offerings easier to scale and buy.
- While fully autonomous AI-driven attacks were not observed for sale, AI is increasingly making discrete elements of attacks cheaper and more efficient.
Conclusion
The Halcyon report shows that AI tools have become widely available on underground markets and that sellers have applied commercial sales models to distribute them. This lowers the technical threshold for various cybercriminal activities and may increase the pool of people capable of carrying out harmful online attacks, even if fully autonomous AI-enabled cyberattacks have not yet been documented for sale.
