An anonymous developer launched an artificial intelligence (AI) chatbot on the dark web in an effort to provide “uncensored” access to information.
The service, known as OnionGPT, was announced on Dec. 18 in a post to a dark web forum by an individual using the pseudonym “root.”
“Say hello to OnionGPT alpha,” root said. “It’s an uncensored smart assistant accessible over Tor.”
The Tor network is a privacy-focused system that routes internet traffic through multiple encrypted servers to hide users’ identities and locations. OnionGPT is hosted on a Tor onion service: a website that can only be accessed through Tor, offering anonymous hosting by hiding the server’s IP address.
Highlighting his goal of unrestricted content, root included in his post simple step-by-step instructions from OnionGPT for manufacturing methamphetamine.
During an encrypted chat with Daily Dot, root noted that OnionGPT’s creation was rooted in his dislike of authority figures as well as dissatisfaction with restrictions imposed by mainstream AI services such as OpenAI’s ChatGPT.
“It is of my personal opinion that information should be accessible freely no matter what,” root said, likening the project to a “give me freedom or give me death” philosophy.
OnionGPT, according to root, utilizes a language model called DarkIdol-Llama-3.1-13.3B-Instruct-1.2-Uncensored. The system was allegedly built entirely using self-owned hardware, including a Mac Mini M4 Pro.
The service is much less complex than mainstream AI chatbots. It only provides text output and cannot create images. Only one question can be asked before the page reloads, meaning follow-up inquiries aren’t an option. And given that OnionGPT is hosted on the dark web, responses take longer than more mainstream AI chatbots.
Nevertheless, root says OnionGPT is providing more than 5,000 responses per day and recently passed a half-million visitors. The service, root emphasized, is “uncensored to the best of my abilities,” but has a strict prohibition on generating responses related to minors. The restriction was added after launch when root noticed efforts by some users to produce such content.
AI expert Dr. Felix M. Simon, a Research Associate at the Oxford Internet Institute, said that while an unrestricted chatbot isn’t necessarily great news, its presence on the dark web “will likely severely limit the number of people using it in the first place.”
“Most people will simply not want to go via a Tor browser to access a chatbot that will in all likelihood not be as capable as the more widely available—and more easily accessible—systems,” Dr. Simon told the Daily Dot. “In addition, for dangerous uses of such systems (e.g. making drugs, building bombs) the problem is not so much in finding the information on how to do it but having access to the right material and equipment.”
“Amid the frenzy about AI systems, it is easy to forget that problematic or illegal content is already fairly easy to come by on the internet, especially for those who know where to look,” Dr. Simon added.
Given potential legal issues surrounding OnionGPT, root argued that he would “comply if absolutely required” if law enforcement “in their infinite wisdom decides that OnionGPT is worthy to be shut down.”
Internet culture is chaotic—but we’ll break it down for you in one daily email. Sign up for the Daily Dot’s web_crawlr newsletter here. You’ll get the best (and worst) of the internet straight into your inbox.
