Microsoft backdoor research raises alarm over open weights
A new Microsoft Research paper shows a backdoor can be planted by modifying only a small part of a model's weights, raising concerns that open-weight models are an attack surface, with risks amplified as recursive self-improvement nears.
2026-11-11 ~ 2026-11-11 · 2 related posts
- Backdoor paper stirs fears: open-weight models as attack surface as RSI nears — DimitrisPapail · 2026-11-11
- Dormant-neuron backdoor paper raises alarm: open-weight models are backdoorable by anyone — DimitrisPapail · 2026-11-11