Poster · Sydney
Benign Reinforcement Learning Can Amplify Latent Backdoors
- When
- Thu 10 Dec 2026, 10:00–13:00 AEDT
- Where
- Hall 1-4, ICC Sydney
Exhibition Centre, Level 1 - Session
- Sydney Poster Session 5
- Authors
- Chen Lai, Javier Rando, Nicholas Carlini, Yiming Zhang
- Organisations
- Anthropic, Google DeepMind
- QRS classification
- Reinforcement learning, Language models, Safety, fairness and privacy