大模型对齐与安全

The Gate Is Only as Honest as Its Contracts: ContractGuard for the Contract Layer of Risk-Aware Causal Gating

门仅与其合约一样诚实：面向风险感知因果门控合约层的ContractGuard

Laxmipriya Ganesh Iyer, Rahul Suresh Babu

专题命中 提示注入 ：prompt injection(abstract)

AI总结 针对工具增强型LLM代理的间接提示注入，提出ContractGuard，通过验证合约完整性（而非风险标签）来防御攻击，在基准测试中实现零注入成功率。

Review the Code, Not the Story: A Vision and Protocol for Code-First Peer Review

审查代码，而非故事：代码优先同行评审的愿景与协议

Jienan Chen

专题命中 提示注入 ：prompt injection(abstract)

AI总结 提出代码优先同行评审协议，由AI系统构建环境、执行实验并生成标准化审查包，将评审焦点从叙述转向可执行证据。

Comments 17 pages, vision and protocol paper