(TWN)-Large language models, like OpenAI's ChatGPT, can be used to produce phishing emails at enormous scale, which, according to experts, can be exploited for bad purposes. 

According to a blog post from cybersecurity company SlashNext, the entry hurdle has now been lowered even further with the introduction of an artificial intelligence bot that resembles ChatGPT and can be readily instructed to produce sophisticated malware.

The system, named WormGPT, appears to have been trained primarily on malware data — and, significantly, lacks any safety guardrails, unlike ChatGPT and Google's Bard. As an example of its capabilities, it can be simply encouraged to develop malicious Python software.
It's a foreboding omen of the times. Cybersecurity is already a demanding undertaking, but the introduction of AI is driving the industry into new, risky area. Even if WormGPT isn't going to hijack the world anytime soon, it could be a foreshadowing of things to come.

WormGPT was discovered by SlashNext employees on a hacker site, where the developer has been selling access to the bot since March, promising that it can do "all sorts of illegal stuff."

WormGPT appears to be based on GPT-J, an older big language model from 2021 produced by EleutherAI, a non-profit organization that creates open source AI systems.

According to PCMag, the hacker behind WormGPT is selling access to the malware for $67.44 each month.

SlashNext employees were astounded by the program's ability to generate well-written phishing emails.

"WormGPT created an email that was not only incredibly persuasive, but also strategically astute, demonstrating its potential for sophisticated phishing and BEC attacks."

Despite its mixed reviews, the bot could provide a glimpse into a frightening future of AI-driven cybercrime, making it more difficult than ever to protect our money and data — and potentially creating even more obstacles for the embryonic AI industry.