HAMBURG — In 2026, open-weight AI models possess advanced capabilities nearly on par with proprietary systems from companies such as OpenAI, Google, Anthropic, and xAI. Unlike those models, however, open-weight versions can have their safety guardrails permanently removed and are often run locally without internet access, making their use difficult to monitor.
A recently developed method called “abliteration” allows people to remove an AI model’s ability to refuse harmful requests by adjusting its underlying parameters, or weights. The tool Heretic automates this process, requiring only two lines of code and completing the task in minutes. According to research by the National Counterterrorism Innovation, Technology, and Education Center (NCITE), Hugging Face now hosts over 6,000 abliterated models—up from about 600 in 2024—and they outnumber models modified by other methods.
Heretic has grown more popular on GitHub since February, according to research by AI security firm Alice. In late April, House lawmakers attended an NCITE demonstration of these models. “What was frightening about this demonstration was how readily available some of this content or software is on kind of the black market right now, and how it can be weaponized and used to manipulate people, destroy lives and build weapons of mass destruction,” said Rep. Andy Ogles (R-TN).
Security researchers have documented real-world misuse. The Counter Extremism Project reported that an individual in a pro-ISIS chat room claimed to use an “uncensored” AI to research the amount and type of explosives needed to destroy Trump Tower in the U.S. On a cybercrime forum, a user seeking ways to bypass AI guardrails for scam calls was directed to Heretic, per Alice’s research. Several accounts on X have also reported using abliterated models to generate pornography.
“It’s jarring when you see it in real time, this sort of bubbly persona with some of the abliterated models that’s like, ‘Oh, what a great idea to create this bomb,’” said Samuel Hunter, senior scientist and director of academic research at NCITE. “Imagine somebody that has no other kind of social connection and it starts to take them down a darker path and really encourage them,” he added.
Noam Schwartz, CEO of Alice, acknowledged risks but emphasized legitimate applications, such as cybersecurity research and catching bad actors. “Everybody can download and operate their own state-of-the-art model and use it for great things and terrible things,” Schwartz said. He noted the accessibility shift: “That was the job of the data scientist, you know, a senior employee. Now, everybody with access to the internet and a laptop for like 400 bucks can actually run this thing on their own machine.”