Heretic tool automates AI model decensoring without post-training

By PulseAugur Editorial · [1 sources] · 2026-09-09 08:03

Heretic, an open-source tool, automates the removal of safety alignment from language models without requiring costly post-training. It utilizes Optuna for parameter optimization, co-minimizing model refusals and KL divergence to maintain intelligence while stripping censorship. This allows developers to decensor transformer models directly from the command line, bypassing manual vector tuning. AI

IMPACT Enables developers to bypass safety alignment in AI models, potentially impacting content moderation and AI behavior.

RANK_REASON The item describes a tool for modifying AI models, not a core AI release or research paper.

Read on Mastodon — mastodon.social →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Heretic tool automates AI model decensoring without post-training

How we ranked this

Signal score: 7 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; The item describes a tool for modifying AI models, not a core AI release or research paper.
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: model release, product, other; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

Mastodon — mastodon.social TIER_1 English(EN) · latreon · 2026-09-09 08:03

Without expensive post-training, Heretic automatically removes safety alignment from language models using Optuna parameter optimization. It co-minimizes model

Without expensive post-training, Heretic automatically removes safety alignment from language models using Optuna parameter optimization. It co-minimizes model refusals and KL divergence to preserve intelligence while stripping censorship. It lets developers decensor transformer …

LINKS github.com/…/heretic

How we ranked this

COVERAGE [1]

Without expensive post-training, Heretic automatically removes safety alignment from language models using Optuna parameter optimization. It co-minimizes model

RELATED ENTITIES

RELATED TOPICS