EPISODE · Jun 9, 2026 · 11 MIN
AI Agent 论文播报 06-08：给 Agent 补上可验证的中间层
from 周六9点半
这一期我们聊一个共同主题：机制化。安全、评测、训练三条线今天都在做同一件事——给 Agent 的中间过程加上可机器验证的契约，不再依赖大模型自己说"对还是错"。本期重点* 恶意 Skill 运行时基准（MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills）：针对 Claude Code、Gemini CLI 这类编码 Agent 的第三方 skill 供应链风险，把攻击空间拆成攻击向量×行为×插入策略 108 个格子，每个恶意样本必须在 Docker 沙箱里真触发系统调用才入库；实测代码注入好造也好抓，提示词注入则攻防同时崩——这是今天最值得收藏的一把"尺子"。* Lean4 工作流形式化验证（Lean4Agent: Formal Modeling and Verification for Agen...去小宇宙查看完整单集简介在小宇宙查看该单集文稿
Embed this episode
Ready to play
AI Agent 论文播报 06-08：给 Agent 补上可验证的中间层
0:00
11:28
1×
No transcript for this episode yet
Similar Episodes
No similar episodes found.
Similar Podcasts
No similar podcasts found.
Frequently Asked Questions
How long is this episode of 周六9点半?
This episode is 11 minutes long.
When was this 周六9点半 episode published?
This episode was published on June 9, 2026.
Can I download this 周六9点半 episode?
Yes. Use the download control on the episode player to save the publisher-provided media file.
URL copied to clipboard!