POLYFLOW: A Neuro-Symbolic Framework for Static Cross-Language Information Flow Analysis
现代软件系统通常由多种编程语言混合构建（如 Python 与 C、Java 与 C 的交互），这种跨语言架构会因语言间的语义差异和隐式交互引入难以察觉的漏洞。传统静态分析器受限于不同语言语义的异构性，往往无法有效建模跨语言的数据流；动态分析
2026-09-02 归档
今日共收录 1494 条安全情报，包含 600 个 CVE，479 篇安全通告，以及 120 篇研究论文。
← 返回归档列表A Formal Analysis of Agent Payment Protocols
该论文针对AI代理支付协议这一新兴的自主商业交易层进行了系统性的形式化安全分析。研究背景在于，AI代理正被授权代表用户购买商品与服务并执行支付，而与传统支付流程不同，此类协议将用户意图、委托授权、凭据使用、结算和履约分散到多个参与方与多个协
AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes
本文针对商用LLM API中模型身份被悄悄替换（如用较小模型替代、量化或包装）导致的不透明现象，提出了一种新的审计方法。现有审计大多基于文本输出通道来判断模型身份，但在代理型（agentic）API场景中，由于模型调用工具时服务堆栈（如Op
OASIS: Optimizing Attacker Sequences for Hard-Label Black-Box Text Attacks
本文介绍了一种名为 OASIS 的方法，旨在优化硬标签黑盒文本攻击中的攻击者序列。传统上，不同攻击方法遵循不同的搜索轨迹，且各自在不同样本子集上取得成功，但已有硬标签黑盒攻击主要关注改进单个攻击者，或通过手动方式简单组合多个攻击者。OASI
HiveTraceGuard-Pro: A Compact Generative Guardrail for Prompt Injection, Jailbreaks, and Adversarial Obfuscation
论文提出 HiveTraceGuard-Pro，一种专为生产环境 LLM 设计的紧凑生成式安全护栏，用于拦截提示注入、越狱和对抗性混淆攻击。作者指出现有公开资料很少涉及俄语提示注入及俄语表层混淆，因此该工作重点面向俄英双语输入。模型基于 Q
抢先加入AI时代顶尖安全团队！阿里云2027届实习生招聘来了！
该信息来源为阿里云先知社区，标题为『抢先加入AI时代顶尖安全团队！阿里云2027届实习生招聘来了！』，发布时间显示为2026-09-02。由于本次输入仅包含标题与来源标签，未提供正文内容，因此无法获取该招聘公告的详细信息。从标题判断，内容属
league/commonmark: Denial of service via distinctly-named attributes in the Attributes extension
### Impact `AttributesExtension` ships with the library but must be explicitly registered on the `Environment`; it is n
league/commonmark: Denial of service in the SmartPunct and Attributes extensions
### Impact Two first-party extensions contain quadratic parsing paths. Both ship with the library but must be explicitl
tornado: multipart split() creates huge temp list before max_parts check -> memory amplification DoS (httputil.py:34)
## Description ### Summary `parse_multipart_form_data` (httputil.py:34) calls `data.split(b"--"+boundary+b"\r\n")` **b
Tornado: Incomplete fix for CVE-2026-35536: cookie attribute injection re-opened via the legacy case-insensitive `**kwargs` path in `set_cookie`
## Summary The CVE-2026-35536 fix added a validation loop that rejects `[\x00-\x20\x3b\x7f]`, but only for the hardcoded
pnpm: Environment secrets exfiltrated via env-placeholder expansion in proxy settings read from an untrusted pnpm-workspace.yaml
## Summary pnpm expands `${VAR}` environment placeholders in the `httpProxy` / `httpsProxy` / `noProxy` settings read f
pnpm: pacquet trust-lockfile install can create dependency symlinks outside the project
## Summary A crafted lockfile alias could reach several install-time filesystem joins. With `--trust-lockfile` or a fro
MLFLOW_ALLOW_PICKLE_DESERIALIZATION=False safety control bypassed by mlflow.statsmodels flavor — RCE via crafted model artifact
## Summary MLflow introduced `MLFLOW_ALLOW_PICKLE_DESERIALIZATION` as a security control to prevent unsafe `pickle.load
[原创] 别傻了，你的引擎加密挡不住"不讲武德"的邪修 —— XA 内存盲打补丁，记偏移秒改的实战套路
[原创] 别傻了，你的引擎加密挡不住"不讲武德"的邪修 —— XA 内存盲打补丁，记偏移秒改的实战套路
AGFlow 三洞审计：补丁追着漏洞跑，有个版本掉队了
AGFlow 三洞审计：补丁追着漏洞跑，有个版本掉队了
Windows11 0day内核提权漏洞分析与利用（CVE-2026-62737）
CVE-2026-62737
Windows11 0day内核提权漏洞分析与利用（CVE-2026-62737）
CVE-2026-73653：Vitest Browser Mode 权限绕过漏洞原理与代码分析
CVE-2026-73653
CVE-2026-73653：Vitest Browser Mode 权限绕过漏洞原理与代码分析
PaperCut Exploitation Escalates to Active Intrusions
CVE-2026-82078 · CVE-2026-81578
CISA has added the vulnerabilities tracked as CVE-2026-82078 and CVE-2026-81578 to its KEV catalog. The post PaperCut Ex
Attackers Exploit Critical JFrog Artifactory Flaw to Mint Admin Tokens Days After Disclosure
CVE-2026-82329
Threat actors are exploiting a newly patched critical security flaw impacting JFrog Artifactory merely days after public
Attackers Exploit Critical Langflow and Rails Flaws in Credential-Probing and C2 Activity
CVE-2026-0768 · CVE-2026-66066
Threat actors are exploiting two critical flaws impacting Langflow and Ruby on Rails, according to new findings from Vul