Posts of last few hours
Please support the site operations by clicking ads.
Why AI Agents With Shell Access Need Sandboxes, Managed Policy and Human Oversight
AI coding harnesses can read files, run shell commands, install software and push code with the authority of whoever launched them. Security teams need enforced sandboxes, managed permissions, short-lived credentials and auditable records before the tools spread across the enterprise.
AI coding harnesses can read files, run shell commands, install software and push code with the authority of whoever launched them. Security teams need enforced sandboxes, managed permissions, short-lived credentials and auditable records before the tools spread across the enterprise.
https://ransomware.databreachtoday.com/blogs/securing-ai-coding-harnesses-p-4228
Unrevoked Credential Exposed in May Enabled Repository Compromise in October
An attacker exploited an unrevoked LiteLLM master key to steal GitHub credentials, bypass release controls and publish a credential-stealing PyPI package downloaded roughly 50,000 times, exposing developer systems, cloud environments and AI provider secrets.
An attacker exploited an unrevoked LiteLLM master key to steal GitHub credentials, bypass release controls and publish a credential-stealing PyPI package downloaded roughly 50,000 times, exposing developer systems, cloud environments and AI provider secrets.
https://ransomware.databreachtoday.com/exposed-litellm-key-enabled-pypi-supply-chain-attack-a-33211
Agency Vows 'Aggressive, Anticipatory' Stance After Losing a Third of Its StaffThe U.S.
Cybersecurity and Infrastructure Security Agency's new CISA 2030 strategy vows to take an "aggressive, anticipatory stance" against adversaries - but analysts say it ignores ransomware payments, liability and the deep staffing cuts that could keep it from delivering.
Cybersecurity and Infrastructure Security Agency's new CISA 2030 strategy vows to take an "aggressive, anticipatory stance" against adversaries - but analysts say it ignores ransomware payments, liability and the deep staffing cuts that could keep it from delivering.
https://ransomware.databreachtoday.com/cisa-2030-analysts-question-agencys-ability-to-deliver-a-33213
HumanX 2026: Trust, Governance and Business Value Define AI's Next Chapter
As AI agents gain autonomy, enterprises face mounting challenges around security, governance and accountability. Drawing on interviews with technology and business leaders at HumanX 2026, ISMG's Pulse Report examines how organizations can balance AI innovation with control, trust and business value.
As AI agents gain autonomy, enterprises face mounting challenges around security, governance and accountability. Drawing on interviews with technology and business leaders at HumanX 2026, ISMG's Pulse Report examines how organizations can balance AI innovation with control, trust and business value.
https://ransomware.databreachtoday.com/ismg-pulse-report-agentic-ai-meets-enterprise-control-a-33214
Blue Initiative Will List Vulnerabilities Affecting Maritime IT or OT Systems
A new public access database launched Monday will catalogue IT and industrial control systems software vulnerabilities that affect maritime systems. The vision "is to have a public and open single source of truth for maritime-related vulnerabilities."
A new public access database launched Monday will catalogue IT and industrial control systems software vulnerabilities that affect maritime systems. The vision "is to have a public and open single source of truth for maritime-related vulnerabilities."
https://ransomware.databreachtoday.com/maritime-sector-to-catalog-nautical-software-vulnerabilities-a-33216
国内唯一！
https://mp.weixin.qq.com/s?__biz=MzI5NjA0NjI5MQ==&mid=2650188533&idx=1&sn=937df72f8b759bf95cfba098e3834108
The defect in a widely distributed open-source library provoked an emergency response, yet exploitation hasn’t been observed in the wild.
The post Experts temper emergency-level alarms over critical Brotli defect appeared first on CyberScoop.
https://cyberscoop.com/brotli-critical-vulnerability-decompression-library-node2shell-broccoli2shell/
A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests. Here is how we addressed the issue.
https://blog.cloudflare.com/kv-cache-exposure-remediation/
MeshCentral — A complete web-based remote monitoring and management web site. Once setup you can ins
https://buaq.net/go-450955.html
它不是内存安全漏洞，也不是逻辑越权。它的本质是：地址解析层中存在三条相互独立的超线性（近二次）性能路径与一个参数上限触发的异常路径，而地址解析在 Node.js 的单线程事件循环上同步执行。
https://xz.aliyun.com/news/92860
本文记录了针对已归档 Python 包 langchain-experimental 的代码审计过程。在该包的 PythonAstREPLTool 组件中，发现其对大模型（LLM）输出内容的过滤机制较弱，且直接使用 exec/eval 执行输入。结合平台判定与最新下载量数据，本文客观分析了该组件在现阶段（已停止维护）存在的安全风险及修复建议。
https://xz.aliyun.com/news/92868
@tiptap/core >=3.7.0,<3.30.5 的 Markdown 属性解析里存在两条二次时间复杂度路径：Pandoc 风格的块属性解析器（attributeUtils.ts）在键值匹配和清理两个阶段各执行一次同型的未锚定贪婪搜索；行内 shortcode 解析器（createInlineMarkdownSpec.ts，以及 Mention 扩展自带的内联实现）对无等号的长词串从每个位
https://xz.aliyun.com/news/92870
OpenTelemetry-Go 的 OTLP 日志 gRPC 导出器（go.opentelemetry.io/otel/exporters/otlp/otlplog/otlploggrpc，受影响区间 v0.5.0–v0.20.0，官方口径 < v0.21.0）会把 OTEL_EXPORTER_OTLP_LOGS_CERTIFICATE 这组环境变量完整解析进 cfg.tlsCfg——根 CA、
https://xz.aliyun.com/news/92871
随机 tcache_key 拦截 double-free，safe-linking 隐藏链表指针，malloc/free hook 被连根移除。
https://xz.aliyun.com/news/92890
CVE-2026-88975（CVSS 3.1 7.5，CWE-400）是 http4s Ember HTTP/2 实现的未认证资源消耗漏洞：读循环在缓冲完整 9 + length 字节之前不做任何帧大小校验，对端可以在 9 字节帧头里声明一个接近 24 位上限的单帧长度，并持续发送接近 16 MiB、但仍不足额的负载，使当前未完成帧的累积缓冲 acc 驻留约 16 MiB——约为 16 KiB
https://xz.aliyun.com/news/92896
一个 xn-- 前缀加一个尾 -，黑名单里的域名就能原样通过校验；ToUnicode("xn--example-.com") 解码出 example.com 却不带错误。
https://xz.aliyun.com/news/92916
过去做大模型安全时，我们习惯把注意力放在模型“看到了什么”：Prompt Injection、RAG 文档投毒、工具返回污染，本质上都在影响模型接收到的上下文。Structured Output、JSON Schema、GBNF、Guided Decoding 这类机制出现后，有一个问题又值得我们关注：模型下一步究竟被允许生成什么？
https://xz.aliyun.com/news/92965
覆盖 11 个分类。Web 方向的主线是解析器差异——三个幽灵的 Java char 截断、Auto-3 的 requestURI 与 Tomcat 归一化不一致、推推字母的零参函数返回值通道，攻击成立的共同前提是程序里存在两个观察同一输入的代码位置而它们看到的字节序列不同。Crypto 方向集中在素数生成环节：Pell 数列的代数关联、p−1 的光滑结构、高位信息的部分泄露、平方根歧义被接口消解
https://xz.aliyun.com/news/92993
一条 8 MiB 的字符串，不需要账号、不需要权限，远程就能让进程 SIGSEGV。glibc 里活了二十四年的 wordexp，栽在一个 ~ 上。
https://xz.aliyun.com/news/92923
Latest Blog Posts
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours ago
- 7 hours 42 minutes ago
- 7 hours 42 minutes ago