1
Give your AI sub-agent a budget, not your keys
给AI子代理设预算和时限，别直接交出钥匙，安全又省心
Spawn a sub-agent in CrewAI, LangGraph, AutoGen, or a Claude sub-agent setup and check what it actually holds: your credentials. The parent's keys, at…
2
Bounded Agents: Delegation Security for Multi-Agent AI Systems
多智能体AI系统的安全命门：静态权限与逐请求评估，如何引发越权风险？
arXiv:2608.15888v1 Announce Type: new Abstract: LLM-based agents can act on behalf of a user to access cloud services, call tools, or invoke agents. A…
3
Model-Generated GitHub Actions Need a Permission Gate
AI自动生成GitHub Actions虽便利，但需权限门防止密钥泄露，开发者必看的安全实践。
A model can write a valid GitHub Actions workflow that still leaks a secret. Lint the syntax, then gate the permissions. The failure mode Most checks …
4
Your Custom AI App Is the New Security Perimeter: Why RAG and Internal Chatbots Need Real Guardrails
为RAG应用和内部聊天机器人提供实时防护，拦截提示注入和越权访问，守护AI安全边界。
Building a custom AI application has become remarkably straightforward. Engineering teams can connect a Large Language Model to internal company knowl…
5
谷歌收紧 Chrome 浏览器通知权限：“干扰”网站每分钟最多推 1000 条
Chrome 新版通知权限自动收紧，智能撤销不常互动网站的推送，通知量最多降三成，有效减少打扰。
IT之家 8 月 12 日消息，谷歌公司昨日（8 月 11 日）发布博文， 披露称针对网站滥用推送通知，已在 Chrome 浏览器中部署多层防御体系。 在谷歌 Chrome 浏览器中，推送通知主要用于新闻提醒、配送更新或价格变动提醒等场景，开发者可借此实时触达用户。不过谷歌表示越来越多的恶意行为者滥…
6
Learn Why Agent Tool Boundaries Fail by Building a Tiny Permission Harness
60行代码亲手复现AI智能体越权瞬间，用真实模型验证工具边界为何失守，比空谈观点更有说服力。
Here is the output this project produces before you read another word: === tool-boundary harness === [PASS] benign-summary model called: read_file pol…
7
苹果曝出数据管理漏洞：部分员工离职后仍可通过个人 iCloud 访问公司机密文件
企业级在线文档工具，统一管理共享文件，支持细粒度权限设置，员工离职后自动移除访问权。
IT之家 8 月 3 日消息，苹果公司起诉 OpenAI 窃取商业机密一案持续发酵。据 The Information 报道，多名前苹果员工透露，即便苹果已切断他们的企业内网权限，他们仍能访问在职期间的共享文档。部分员工甚至在文件更新时，还能继续收到通知。 这些文档包含产品发布会计划等机密资料。前员…
8
HUQAN: The Deterministic Trust Layer That Tells AI Agents "Wait, I Decide First"
AI代理乱跑谁叫停？这个确定性信任层让“人类拍板”成为铁律。
We gave agents tools — but who gets to say "no"? Over the last year or two, the agent ecosystem has grown incredibly fast: LLM-based agents that touch…
9
Human-in-the-Loop Agentic DevOps: Govern AI Automation in GitHub Issues
AI代理自动化GitHub Issues时权限如何控？人机协作治理方案来了！
Human-in-the-Loop Agentic DevOps: Govern AI Automation in GitHub Issues The biggest question in agentic DevOps is no longer, "Can an AI agent automate…
10
Privilege and confidentiality in generative AI workflows
生成式AI工作流如何保护特权和机密？专门探讨其中的权限与保密挑战，适合AI安全与隐私关注者。
arXiv:2607.05479v1 Announce Type: cross Abstract: Generative AI (GenAI) systems store and process client data in three distinct ways: in the model's p…
11
Access Is Not Agency
像控制AI代理的访问权限与行动边界一样，Flowise让你精细定义agent的能力范围，确保安全与效率
Your agent can read Slack. It can search email. It can query the CRM, open GitHub issues, check billing, browse docs, edit a spreadsheet, draft a cust…
12
Show HN: Legant gives AI agents bounded authority to act on your behalf
给 AI 代理划清权限边界，放心让它替你办事的开源方案。
Article URL: https://github.com/legant-dev/legant Comments URL: https://news.ycombinator.com/item?id=48645063 Points: 1 # Comments: 0
13
提升流畅度：谷歌安卓 17 给 App 内存占用套上“紧箍咒”
谷歌安卓17强制App内存上限并新增权限控制，提升系统流畅度但引入Wi-Fi断网问题，开发者与用户需关注。
IT之家 6 月 20 日消息，科技媒体 Android Authority 昨日（6 月 19 日）发布博文， 指出在安卓 17 系统更新中，谷歌为 Pixel 10 等设备带来 4 项隐藏改进，包括自动终止内存占用过高 App 等。 一、应用内存限制 在安卓 17 系统中，谷歌引入了应用内存限制…
14
Show HN: Coding Tools MCP – Give any LLM agent the ability to code
为LLM代理打造的开源编码工具集，20个实战工具，权限管控与工作区隔离，让AI自动修复测试失败。
Article URL: https://github.com/xyTom/coding-tools-mcp Comments URL: https://news.ycombinator.com/item?id=48594594 Points: 2 # Comments: 0
15
AI Agents Don't Need More Memory. They Need Governed Recall.
这篇文章颠覆了AI代理记忆的主流认知，指出关键不在于容量，而在于权限和语义检索的受控召回。
Most AI Agent Memory discussions start from the same assumption: If the agent forgets, give it more memory. More chat history. More retrieved document…
16
Outpost – Capability-based credential proxy for AI agents (Hermes, Openclaw)
基于能力的安全凭证代理，让AI代理不再直接持有原始API密钥，降低泄露风险
Article URL: https://github.com/sausin/outpost Comments URL: https://news.ycombinator.com/item?id=48554287 Points: 1 # Comments: 0
17
GatekeeperAI – self-hosted governance platform for AI apps your team is building
为团队AI应用提供自托管治理，保障安全合规，开源可自定义策略。
Article URL: https://github.com/jacobthomasmichael/GatekeeperAI/blob/main/README.md Comments URL: https://news.ycombinator.com/item?id=48508622 Points…
18
The ways we contain Claude across products
Anthropic披露Claude产品中从零信任到高权限的工程实践，平衡效率与安全。
Article URL: https://www.anthropic.com/engineering/how-we-contain-claude Comments URL: https://news.ycombinator.com/item?id=48392082 Points: 223 # Com…
19
Jo – Secure Programming for the AI Era
AI时代编程安全新范式：Jo语言从设计上解决AI生成代码的权限失控问题。
Article URL: https://jo-lang.org/blog/2026-06-04-introducing-jo.html Comments URL: https://news.ycombinator.com/item?id=48406714 Points: 3 # Comments:…
20
GitLab发布紧急安全更新 修复高危XSS与API拒绝服务漏洞
GitLab紧急修复四项高危漏洞，XSS与API拒绝服务风险需立即升级