A recent study by Indiana University reveals how illegal AI models thrive on the black market, generating substantial profits. These large language models (LLMs) present a growing risk to cybersecurity, creating new opportunities for malicious actors. With ChatGPT’s 200 million weekly active users driving OpenAI’s value to $100 billion, illegal models are making a name for themselves in underground circles.

Illegal AI Models: A Growing Threat

Illegal large language models can earn up to $28,000 in just two months on the “grey” market, according to a study published on arXiv, managed by Cornell University. The study examined over 200 examples of malicious LLMs (Malicious LLM or MALA) listed on illegal markets between April and October 2023. These models typically fall into two categories: uncensored models built on open-source platforms and jailbreak models that bypass commercial restrictions.

“We must study these models now to prevent widespread damage,” says Xiaofeng Wang, professor at Indiana University Bloomington and co-author of the study. “Our goal is to stay ahead of attackers before they cause harm.”

AI Models and Their Impact on the Black Market

While some hackers bypass restrictions in commercial LLMs, this is uncommon. Instead, illegal AI models have emerged to meet the demand on underground markets, where creators profit from selling them.

Zilong Lin, a researcher at Indiana University, states, “Most MALA services on underground forums exist primarily to generate income.” These LLMs are used for malicious purposes, such as writing phishing emails and developing malware for cyberattacks. According to a study on cybersecurity, LLMs can reduce the cost of producing phishing emails by 96%.

Potent Tools for Cybercrime

The capabilities of black-market AI models vary, but some are especially dangerous. Lin and Wang identified two uncensored LLMs, DarkGPT and Escape GPT, that successfully generated accurate code in two-thirds of cases. The code often bypasses antivirus tools, increasing the risk of successful cyberattacks.

Another malicious LLM, WolfGPT, priced at $150 for access, creates phishing emails that evade most spam filters, making it a powerful tool in the hands of cybercriminals.