A startup called Abliteration AI removes refusals from free AI models and offers them to anyone, which helps some users and raises real safety worries.
A company called Abliteration AI takes powerful free-to-download AI models, removes the rules that make them refuse certain requests, and offers the result to anyone. Its founder, a 36-year-old Apple engineer who goes by Thomas, says he wants people to set their own limits. Many in the AI world find the idea close to blasphemy.
Thomas began in November 2025. He disliked sending information to the big AI labs, and he was annoyed that their models could change from one day to the next. They also turned down useful work. When he tried to study logs of cyber attacks, or to find which app users might pay for a new feature, his requests were blocked and his account was banned.
So he built a $20,000 cluster of computers in his apartment to run free models at home. Those models do not share anything with their makers. But they still refused many requests, because they come with safety rules built in.
Free models of this kind are called open-weight. Online communities on Reddit, Discord and X share ways to tweak a model's inner settings so its refusals disappear. Thomas also tried Dolphin, an uncensored model made by Eric Hartford of Charlotte, North Carolina, starting from Meta's Llama.
Critics add a business angle. OpenAI and Anthropic both charge extra for models with fewer cybersecurity limits, and their chatbots balk at requests to generate data for training another model.
The danger is plain. A model without guardrails will gladly help someone start a meth lab, build a bomb or hack a major corporation. Many AI safety researchers see these models as very dangerous. Recent incidents in which OpenAI and Anthropic agents went rogue involved models with fewer cybersecurity limits.
Thomas and others admit the risk. They compare it to a hammer, which can hurt people if misused. In his words: "It's not about it being unrestricted; it's about you setting the guardrails."
The material does not say how many people use Abliteration AI, what it charges, or what checks it puts on customers.
Before using any AI tool, find out who runs it and where your data goes, because less restricted models cut both ways.
Cognizant is putting 25,000+ staff on Claude and will keep running the AI after launch. That matters if your business wants AI that someone stays accountable for.
Read more: Cognizant Will Run Claude AI Systems Inside Client CompaniesA report says Fortune 500 firms spend under 0.1% of their costs on Claude, a sign that big companies are still early with AI and that you are not behind.
Read more: Top US companies spend under 0.1% of their costs on Claude AIOpenAI Academy is offering free, hands-on ChatGPT classes for caregivers in five US cities, aimed at saving time on everyday tasks.
Read more: Free ChatGPT workshops are coming to caregivers in five US cities
