AI Papers Daily

#Guardrail

1 篇

LiSA: Lifelong Safety Adaptation via Conservative Policy Induction

LiSA：通过保守策略归纳实现终身安全适配

🧭 Agent ★ 深度解读

Google Cloud AI Research · Seoul National University · Google

在稀疏、嘈杂反馈下，通过结构化策略记忆与证据感知重用的保守归纳，持续改进AI护栏

Agent | 2026-05-13 | HF Daily | score 8/10

GuardrailPolicy InductionMemoryLifelong LearningAI AgentsSafety