SyncAI.news, a Varaisys broadcasting
Tear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers
WK

Will Knight

· 1 min read

BusinessWIRED: AI

Tear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers

Devon Thomas hardly seems like a renegade. Members of his bible study group, he tells me, would probably describe him as unfailingly kind and polite.

And yet Thomas is doing something that many in the AI world consider borderline blasphemous. His new company, Abliteration AI, removes safeguards from the world’s most powerful open-weight models. It also makes these models available to anyone.

As someone who covers AI and constantly hears about the need for increased safety, I was fascinated to learn about this subversive business venture. I reached out to Thomas, and we began corresponding by text and chatting on the phone. What I learned challenged my assumptions about the AI industry’s safety narrative—namely, that uncensored models are primarily the domain of criminals.

Thomas is 36 years old, with a thick beard, unpretentious glasses, and a self-deprecating laugh. He grew up in Indiana and recalls splitting his time between sports and the computer lab, where he spent hours using Netscape Navigator to explore the early web. He studied computer science at Indiana University, Bloomington, and joined Apple in 2020.

Thomas’s AI rebellion began in November 2025. His day job as an engineer at Apple involved working with AI, but outside of work he disliked using models hosted by the big labs; it meant potentially sharing information with them, and Thomas was frustrated that the models could change from one day to the next. Even more irksome to Thomas, these models would often refuse to do useful things when he was prototyping new ideas, automatically rejecting inputs that included any user information. He found his requests blocked and his account even banned when he tried using a model to analyze the logs of cyber attacks or to identify which users of an app might be most likely to pay for a new feature.

After trying Dolphin and other uncensored models, Thomas found a simple way to create his own—a technique known as abliteration.

Original source

This story was published by WIRED: AI and written by Will Knight. SyncAI.news shows a preview; the complete article is on the publisher's site.

Read the full story on wired.com

Similar News