Anthropic finds clean contrasting data can strengthen poisoned fine-tuning backdoors
thione · x · 2026-11-17
In controlled experiments on open-weight models, Anthropic found that adding contrasting clean data to poisoned fine-tuning can strengthen the hidden backdoor effect rather than dilute it — challenging the intuition that clean data washes out poisoning, with direct implications for open-weight supply-chain security.
More from Safety
- United Airlines bans third-party AI agents from booking flights, citing security and federal rules — thione · 2026-11-17
- Singapore sought US help on Nvidia chip-diversion probe; sources say Washington offered little aid — thione · 2026-11-17
- OpenAI deploys Control Room infra that can halt model outputs within 35 seconds — thione · 2026-11-17
- Fireworks AI confirms hackers accessed billing info, some customer datasets and partial model weights — thione · 2026-11-17
- France unveils Paris Call, a voluntary frontier-AI safety commitment — thione · 2026-11-17
- California sues OpenAI over AI model loss of control, citing Hugging Face and others — Typical-Code-7006 · 2026-11-17