1
微软 CEO 纳德拉：支持 AI 审慎发展与独立审计，AI 行为准则明日公布
IT之家 9 月 14 日消息，微软首席执行官萨蒂亚 · 纳德拉今天在 X 平台发文，谈及微软的前沿 AI 发展理念。 纳德拉表示，任何对超级智能的探索都必须在核心原则约束下运行：如果人们构建的 AI 无法帮助人类，且无法被人类控制，那么它就不值得发展。他认为 AI 需要加速发展，使其成果能够得到广…
2
The State of AI Disclosure 2026: what 1,088 EU sites' chat widgets say
一键扫描欧盟网站AI聊天部件的披露情况，助你快速评估合规风险与透明度。
Article URL: https://disclosureproof.com/research/state-of-ai-disclosure/ Comments URL: https://news.ycombinator.com/item?id=49440371 Points: 1 # Comm…
3
Write-Side Custody: Trust Begins Before Memory
AI记忆栈关键一环：在写入端建立信任，让事后审计无懈可击，深度解析写侧托管机制。
Part 5 of the Building the AI Memory Stack series The previous articles introduced the Reasoning Ledger and then worked through what a single ledger r…
4
Hidden in the Request: Explaining Unethical LLM Compliance through Token Relevance
用Token相关性拆解大模型为何会答应不道德请求，为AI安全审计提供新视角。
arXiv:2608.23264v1 Announce Type: new Abstract: Although Large Language Models (LLMs) are aligned to optimize for both helpfulness and harmlessness, t…
5
Beyond Endpoint Gains: A Weight-Delta Audit of Medical Specialization
模型在医学专科化时权重如何变化？这份审计报告揭示端点表现之外的深层机制。
arXiv:2608.20768v1 Announce Type: new Abstract: Specialist language models are usually understood through endpoint gains: the generalist scores lower,…
6
Is Your AI Account Hacked? Quick Signs & Fixes
AI账户被黑？用这份检查清单快速发现入侵迹象，清理多余管理员权限，守护你的AI资产。
Photo by Steve A Johnson on Unsplash TL;DR: Use this concise checklist to spot a compromised AI account, verify the intrusion, and lock down the breac…
7
The LLM rewrote the email after I approved it
AI代理篡改已批准邮件？600万小时运行揭示真正痛点不在模型智力，而在边界失控。
Article URL: https://www.crusbro.com/en/product-architecture.html Comments URL: https://news.ycombinator.com/item?id=49384800 Points: 2 # Comments: 0
8
LLM Reasoning Traces Are Not Audit Records
大模型的思维链并非真实决策的审计日志，揭开推理可解释性的关键盲区。
Article URL: https://rye.ai/blog/cot-faithfulness-reasoning-traces-not-audit-logs/ Comments URL: https://news.ycombinator.com/item?id=49371034 Points:…
9
LEDGER: Claim-to-Evidence Trace Graphs for Auditing LLM Agents
用追溯图把智能体回答与证据链绑定，审计大模型不再黑箱。
arXiv:2608.18398v1 Announce Type: cross Abstract: Large language model (LLM) agents can now carry out long-horizon technical workflows involving compl…
10
Did OpenAI's AI hack Hugging Face? An evidence audit
OpenAI的AI是否攻击了Hugging Face？一份证据审计带你梳理疑点与真相。
Article URL: https://speedyweedyops.org/did-openai-ai-hack-hugging-face/ Comments URL: https://news.ycombinator.com/item?id=49367230 Points: 1 # Comme…
11
Encoded but Not Actionable: Auditing the Decode-Generate-Steer Gap in Frozen LLMs for Geometric Constraints
冻结大模型能“看懂”几何约束却不会用，审计揭示解码-生成-引导三环断裂，值得关注。
arXiv:2608.17843v1 Announce Type: cross Abstract: Large language models (LLMs) have demonstrated strong performance on structured reasoning tasks, but…
12
AI-Generated GitHub Copilot “Autofix” Allowed Compromise of Snowflake's Jira
AI生成的“修复”竟埋下致命漏洞，Copilot误判安全放行，一场针对Snowflake Jira的供应链攻击实录**
Article URL: https://www.wiz.io/blog/red-agent-snowflake-copilot-cicd-bug Comments URL: https://news.ycombinator.com/item?id=49331423 Points: 292 # Co…
13
How to Audit a npm Package for Malware in 10 Minutes
10分钟排查npm包恶意代码，守住环境变量和CI管道不被陌生代码偷走凭据的实用审计流程。
You are about to run npm install some-package and hand a stranger's code full access to your machine, your environment variables, and your CI pipeline…
14
Prior Audit-Repair Context Shifts LLM Verifier Thresholds Toward Leniency
同一个验证器，在“自己修过”的上下文里会变得更宽松，审计流程中的隐性偏斜值得警惕。
arXiv:2608.16003v1 Announce Type: new Abstract: Automated checking pipelines increasingly place one language model as the checker and another (or the …
15
Whose doctor does the AI recommend? An algorithm audit of reputation and demographic signals in large language model-assisted physician choice
AI推荐医生会偏向谁？算法审计揭示大模型中的声誉与人口信号偏差。
arXiv:2608.14399v1 Announce Type: cross Abstract: Patients increasingly ask large language model (LLM) assistants which doctor to see, making these sy…
16
Sentinel Scan: an authorized LLM red-team audit, run by an AI agent
用AI代理跑授权红队审计，15类攻击全拦下，独立LLM当裁判而非关键词过滤，安全评估思路值得一看。
Article URL: https://fbirds5230.github.io/sentinel-scan/ Comments URL: https://news.ycombinator.com/item?id=49314841 Points: 2 # Comments: 0
17
Auditable Agents
让AI智能体行为可追溯、可问责，破解黑箱难题的前沿研究
arXiv:2604.05485v2 Announce Type: replace Abstract: LLM agents call tools, query databases, delegate tasks, and trigger external side effects. Once an…
18
7 Backend Security Mistakes Node.js Developers Should Avoidv
Node.js后端必看：7个易踩的安全坑，从密码存储到密钥泄露，逐一拆解避坑要点。
7 Backend Security Mistakes Node.js Developers Should Avoid 🔐 Building a Node.js backend that works is one thing. Building a backend that remains secu…
19
Do LLM Recommenders Know When They're Hallucinating? Auditing Confidence Calibration in Catalog Faithfulness
揭秘LLM推荐系统幻觉：模型是否自知？联合审计幻觉率与置信度校准，为目录忠实度提供新视角。
arXiv:2608.10008v1 Announce Type: cross Abstract: LLM recommenders for top-$K$ item suggestion regularly emit titles outside the target catalog. Prior…
20
AIO.GEO Protocol
审计AI搜索结构的利器，支持试运行修复与结果验证，适合优化搜索质量。
Audit AI search structure. Dry run fixes. Receipts. Discussion | Link