Meta FAIR's WM-Basin Detects 79% of Distilled LLMs via API Alone

Meta FAIR's new paper introduces WM-Basin, a black-box method that detects distilled LLMs through API access alone by reading watermark traces, achieving a 79% detection rate against API distillation attacks.

2026-10-29 ~ 2026-10-30 · 2 related posts

• Meta FAIR's WM-Basin flags 79% of distilled models at 0.1% false-positive rate — dair_ai · 2026-10-29
• Meta paper: WM-Basin detects 79% of distilled LLMs from watermark traces in weights — rohanpaul_ai · 2026-10-30