Fine-tuning Large Language Models (LLMs) on instruction data can quietly erode their safety, even when no training example looks harmful. Because this safety-degrading data looks benign, standard cont
SimpleLoRIF uses gradient alignment to filter instruction data that silently degrades LLM safety, yielding a safer dataset that reduces adversarial success while preserving capabilities.
Fine-tuning Large Language Models (LLMs) on instruction data can quietly erode their safety, even when no training example looks harmful. Because this safety-degrading data looks benign, standard content filters struggle to catch it. We tackle this problem by developing SimpleLoRIF, a scalable gradient embedding method that identifies fine-tuning examples whose gradients align with unsafe behavior. By simplifying a low-rank attribution method and restricting backpropagation to the final four transformer blocks, SimpleLoRIF can embed datasets on the order of a million examples, boosting gradient attribution from a small-scale analysis tool into a practical filter for full post-training…
Summary and abstract excerpt via Hugging Face Papers ↗ · arXiv.
Follow the Gradient: Tracing Safety Degradation Back to Training Data was first seen on Hugging Face · Daily papers on Nov 2, 2026. Since then it has appeared in 15 checks on 2 different days. Its best position was #6 of 30, reached on Nov 3, 03:28 UTC. It dropped off the list after Nov 3, 23:28 UTC.
Position on Hugging Face · Daily papers at every check (lower is better).
Best rank per day (UTC).
Platform metric as reported by Hugging Face.
Archived daily, weekly and monthly lists that include this entry.
Not at the moment. It was last on the list on Nov 3, 23:28 UTC, after 2 days of trending.
#6 on Hugging Face · Daily papers, on Nov 3, 03:28 UTC. Trendoday checks this list every 120 minutes.
It was first seen on Nov 2, 2026 and has been on the list on 2 days (15 checks) since.