Hobbyists Strip Safety Guardrails From Open AI Models, Raising Misuse Concerns

A group of developers known as "abliterators" is releasing modified AI models that have had their refusal behaviors removed, so they answer requests that mainstream chatbots decline. Proponents argue the models help security researchers strengthen cyberdefenses and give users more freedom from corporate restrictions. Critics warn the same openness could let bad actors obtain dangerous information, such as bomb-making instructions. The modified models are being widely shared across the internet, making the trade-off between access and safety a growing point of debate. The article frames the movement as a contest between user autonomy and the risks of unrestricted AI.