# Open-Weight Safeguards Are Removable

Cheap weight edits keep stripping safety tuning from open-weight models, which pushes governance away from refusal behavior and toward capability thresholds and access controls applied before release.

- Conviction: 40 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-11-12T00:00:00.000Z
- Last updated: 2026-11-12T14:14:59.720Z
- Canonical: https://polylog.news/ai/trends/open-weight-safeguards-removable
- Publisher: Polylog

## Recent evidence

- [confirms] A guide to running Gemma 4 locally with its refusals stripped out has been saved 14,000 times (2026-11-12): A guide to running 'abliterated' Gemma 4 locally has been saved 14,000 times. Hugging Face model cards report refusal rates on harmful prompts falling from 99 in 100 to 26 in 100 through weight edits with no retraining. Widely shared, cheap removal of safety tuning undercuts refusal-based safeguards on a major lab's open release.