Study finds diluted poisoned fine-tuning data can strengthen hidden backdoor effects
Researchers describe 'homeopathic learning,' where poisoned training data with little effect on its own creates a hidden backdoor once combined with clean, contrasting data.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-10 05:00 · arXiv cs.AI
Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effect