Five labels side by side
| Unrestricted AI | Uncensored AI | Abliterated AI | Jailbroken AI | Filter-free AI | |
|---|---|---|---|---|---|
| What it means | A model whose refusal behavior was removed, by any method. | The common label for the same thing, often used for fine-tunes. | Refusals removed by editing the weights. | A normal model talked past its refusals with a prompt. | A host that adds no content filter on top of the model. |
| What changes | Depends on the method | Weights, usually | Weights | Only the prompt | The host |
| Where you use it | API or local | API or local | API or local | Any chat app | API |
| Example | Kimi K3.1 Pro | Hermes 3 405B | GLM-5.3 | None listed | None listed |
| Survives a new chat | Yes | Yes | Yes | No, the prompt must be repeated | Yes |
Where each one acts
Which to look for
Abliterated
The refusal is gone from the weights, so it holds in every chat and every tool.