GitHub · p-e-w/heretic↑ +211 today
Exceed Security Boundaries in Language Models
Heretic is a technical tool that removes security alignment constraints from transformer-based language models without the need for expensive retraining processes. Using the 'Abliteration' technique, this solution is advanced and research-oriented.
Updates
- November 8, 2026: Stars 39,601 → 39,947, latest release v2.1.0 (November 8, 2026).
- November 6, 2026: Stars 37,706 → 39,601, latest release v2.0.0 (November 1, 2026).
- November 2, 2026: Stars 36,044 → 37,706, latest release v2.0.0 (November 1, 2026).
- October 21, 2026: Stars 34,271 → 36,044, latest release v1.4.0 (June 14, 2026).
Installation
Clone official repo
git clone https://github.com/p-e-w/heretic.gitWhat does it do?
- It removes the constraints in the model with the 'Abliteration' technique.
- Does not require expensive retraining.
- The method is open source and automatic.
Responsibility note
This tool removes the model's security protections. Legal and ethical responsibility for the outputs to be produced belongs entirely to the user; Harmful or illegal use is not recommended. For research/advanced purposes only.
Related dictionary terms
Who it is forML researchers / advanced users
DifficultyAdvanced knowledge of working with model weights
What doesRemove security constraints from the model
FeeFree · open source (AGPL-3.0)
LicenseAGPL-3.0 · detail below
License: AGPL-3.0 · free to use; However, if you change the code and distribute it or offer it as a service, you must also open your own source (copyleft).
Links
TreScout did not build this tool · we found it in GitHub trends and wrote it up. This page describes the repository as of 2026-05-28: The star count and our text belong to that day, the repository may have changed since. Check the repository link for the current state. The introductory text was prepared with AI from the repository's own documentation, and this page was machine-translated from the Turkish original · the Turkish version prevails.