Heretic 2.0: 88 million downloads of models without guardrails
Most AI models are trained to refuse some requests. Some users call that safety layer 'censorship' and want it gone. Heretic is a free tool that removes it from open-weight models, the models anyone can download and run on their own computer.
After almost five months of work, its developer announced Heretic 2.0 on r/LocalLLaMA. The numbers are big: 37,000 GitHub stars, more than 8,500 modified models on Hugging Face, and 88 million total downloads. The author says the project now goes beyond its first purpose, toward 'model modification for everyone'.
Some people use such tools for research or to avoid silly refusals. But the same click removes guardrails that labs spend months building.
On Sunday we saw Claude lock more doors. Heretic shows the other side: with open models, anyone can remove the locks. At 88 million downloads, guardrails are a choice for the user, not a promise from the lab.
