Beyond SFT: Reinforcement Learning for Safer Large Reasoning Models with Better Reasoning Ability
By: Jinghan Jia, Nathalie Baracaldo, Sijia Liu
Potential Business Impact:
Makes smart computers think safely and correctly.
Large reasoning models (LRMs) extend large language models by generating explicit chain-of-thought (CoT) reasoning, significantly improving mathematical and logical problem solving. However, this explicit reasoning process also introduces new safety risks, as unsafe behaviors often emerge within intermediate reasoning trajectories, even when final answers appear harmless. Existing safety alignment approaches primarily rely on supervised fine-tuning (SFT) over safety-oriented long CoT datasets. While intuitive, we find that SFT produces inconsistent safety improvements, degrades reasoning ability, and generalizes poorly across model families. These limitations suggest that purely supervised approaches are insufficient for robust safety alignment in LRMs. To address this, we investigate reinforcement learning (RL) as a complementary optimization framework for LRM safety training. Unlike SFT, RL directly optimizes model policies with reward feedback, enabling more adaptive and stable alignment. Extensive experiments across multiple model families and benchmarks show that RL achieves stronger and more consistent safety gains while maintaining reasoning competence. Further analysis of reflection dynamics and token-level entropy reveals that RL suppresses unsafe exploratory reasoning while preserving reflective depth, leading to safer and more reliable reasoning processes.
Similar Papers
Mitigating Forgetting Between Supervised and Reinforcement Learning Yields Stronger Reasoners
Computation and Language
Makes AI smarter by learning from mistakes.
Breaking the Safety-Capability Tradeoff: Reinforcement Learning with Verifiable Rewards Maintains Safety Guardrails in LLMs
Machine Learning (CS)
Trains AI to be smart and safe together.
Beyond Two-Stage Training: Cooperative SFT and RL for LLM Reasoning
Computation and Language
Teaches computers to learn better and faster.