Breaking

Open-Source Tool 'Heretic' Claims to Remove Refusals from Locally Run Language Models Without Retraining

November 3, 2026 at 17:46 EST

Open Source
Security

A free, open-source tool called Heretic has been released that its developers say can modify any language model a user can run on their own hardware, suppressing refusals by default without any additional training.

Open-source release · Heretic

A free tool that strips refusals from locally run language models in one command

Heretic modifies existing models without training or fine-tuning. Its developers say it works on any model a user can run on their own hardware, and it suppresses refusals by default.

Command to install via pip

30–40

Minutes for the full process, per the announcement

Optional traits to reduce: slop, sycophancy, humor

Training or fine-tuning steps required

Reported runtime, shown to scale

Each block = 5 minutes. Hardware used for the estimate was not specified.

30 min

Shorter estimate

40 min

Longer estimate

The full process is roughly one ten-minute gap wide: 30 to 40 minutes, depending on setup.

How Heretic changes a model

Existing model

Any model the user can run locally

→

Heretic applied

No training or fine-tuning involved

→

Refusals suppressed

The default behavior

→

Optional traits

Reduce slop, sycophancy, or humor

Supporters say

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.