AI Security AI安全 2h ago Updated 1h ago 更新于 1小时前 49

Attackers Exploit Critical Langflow and Rails Flaws in Credential-Probing and C2 Activity 攻击者利用Langflow和Rails的关键漏洞进行凭证探测和C2活动

Two critical vulnerabilities are being actively exploited: CVE-2026-0768 (CVSS 9.8) in Langflow, allowing unauthenticated arbitrary Python code execution as root, and CVE-2026-66066 (CVSS 9.5) in Ruby on Rails, enabling file reads, secret leakage, and remote code execution Threat actors are conducting credential harvesting and reconnaissance, targeting environment variables (LANGFLOW_SUPERUSER, OPENAI_API keys, AWS credentials), reading secret keys, and checking SSH access and bash history Attac 攻击者正利用Langflow的CVE-2026-0768(CVSS 9.8)和Ruby on Rails的CVE-2026-66066(CVSS 9.5)进行凭证探测与命令控制活动。 VulnCheck检测到360起攻击事件,主要源自俄罗斯,目标包括英国、新加坡和以色列的蜜罐系统。 攻击者通过上传恶意图片利用Active Storage与libvips的处理差异,实现任意文件读取和远程代码执行。 漏洞利用导致攻击者部署凭证窃取工具、代理和SimpleHelp远程访问软件,并禁用auditd以规避检测。 超过7,100个暴露的Ruby on Rails实例存在风险,AI开发平台因可访问敏感凭证和

72
Hot 热度
68
Quality 质量
70
Impact 影响力

Analysis 深度分析

TL;DR

  • Two critical vulnerabilities are being actively exploited: CVE-2026-0768 (CVSS 9.8) in Langflow, allowing unauthenticated arbitrary Python code execution as root, and CVE-2026-66066 (CVSS 9.5) in Ruby on Rails, enabling file reads, secret leakage, and remote code execution
  • Threat actors are conducting credential harvesting and reconnaissance, targeting environment variables (LANGFLOW_SUPERUSER, OPENAI_API keys, AWS credentials), reading secret keys, and checking SSH access and bash history
  • Attackers are deploying Python credential harvesters, proxy agents, SimpleHelp for remote access, disabling auditd for forensic evasion, and enlisting compromised hosts into XMR cryptocurrency miner botnets
  • Over 7,100 exposed Ruby on Rails instances and thousands of vulnerable Langflow hosts have been identified, with the majority located in the U.S., Germany, Malaysia, Brazil, and India
  • A patched Rails version (8.1.3.1) blocks the libvips file read but does not neutralize the variation-key Marshal deserialization RCE gadget, meaning the vulnerability remains partially exploitable even after patching

Why It Matters

This represents a significant escalation in threat actor interest targeting AI development platforms, which serve as high-value entry points due to their access to sensitive credentials, cloud environments, and connected systems. The partial ineffectiveness of the Rails patch underscores the importance of defense-in-depth strategies, as relying solely on vendor patches may leave organizations exposed to ongoing exploitation.

Technical Details

  • CVE-2026-0768 (Langflow, CVSS 9.8): A user-supplied input validation flaw that allows unauthenticated attackers to execute arbitrary Python code in the context of the root user on affected Langflow deployments
  • CVE-2026-66066 / KindaRails2Shell (Ruby on Rails, CVSS 9.5): Exploits a discrepancy between Active Storage and libvips in how they read input files; requires applications to use libvips for Active Storage image processing and accept untrusted image uploads. Successful exploitation leaks environment variables, secret_key_base, master keys, database passwords, cloud credentials, and API tokens, leading to RCE
  • Attack chain observed: Threat actors exploit CVE-2026-5027 to deploy credential harvesters and proxy agents, disable auditd to create forensic blind spots, exploit CVE-2026-0769 to drop .sysd payloads, pivot to scan for additional targets, and exploit CVE-2025-3248 to enlist hosts into XMR mining botnets
  • Patch limitation: Testing of Rails 8.1.3.1 confirmed that while the libvips file read is blocked, the variation-key Marshal deserialization RCE gadget still executes on patched servers when given a valid signature
  • Detection scale: VulnCheck recorded over 50 detections within hours of the August 30, 2026 disclosure, rising to 360 detections, with source traffic primarily originating from Russia and a single IP address in France establishing C2 to a host in Israel

Industry Insight

  • Organizations running Langflow or Ruby on Rails should immediately audit their deployments for exposure, prioritize patching, and implement network segmentation to limit lateral movement, especially since the Rails patch is incomplete and additional mitigation layers are required
  • AI development platforms should be treated as high-value targets requiring hardened security configurations, including restricted network access, credential isolation, and continuous monitoring for anomalous environment variable queries and unauthorized file access patterns
  • The sustained and sophisticated attack campaigns—spanning credential theft, cryptomining, and persistent backdoor deployment—indicate that threat actors are systematically targeting the AI infrastructure supply chain, prompting organizations to adopt zero-trust architectures and assume breach postures for all externally exposed AI development tools

TL;DR

  • 攻击者正利用Langflow的CVE-2026-0768(CVSS 9.8)和Ruby on Rails的CVE-2026-66066(CVSS 9.5)进行凭证探测与命令控制活动。
  • VulnCheck检测到360起攻击事件,主要源自俄罗斯,目标包括英国、新加坡和以色列的蜜罐系统。
  • 攻击者通过上传恶意图片利用Active Storage与libvips的处理差异,实现任意文件读取和远程代码执行。
  • 漏洞利用导致攻击者部署凭证窃取工具、代理和SimpleHelp远程访问软件,并禁用auditd以规避检测。
  • 超过7,100个暴露的Ruby on Rails实例存在风险,AI开发平台因可访问敏感凭证和云环境而成为新目标。

为什么值得看

这篇文章揭示了AI开发平台和Web框架正成为高级威胁行为者的新目标,攻击者利用关键漏洞进行凭证窃取和远程代码执行,对AI基础设施安全构成直接威胁。了解这些漏洞的利用方式和攻击模式,有助于AI从业者和安全团队及时修补系统、加强监控,并调整安全策略以应对日益增长的针对AI生态系统的网络攻击。

技术解析

CVE-2026-0768是Langflow中的输入验证缺陷,允许攻击者在root上下文中执行任意Python代码,CVSS评分9.8,表明其严重性极高。攻击者可通过构造恶意输入直接控制服务器,获取最高权限。

CVE-2026-66066(代号KindaRails2Shell)利用Ruby on Rails的Active Storage与libvips之间的处理差异,允许未认证攻击者读取任意文件并泄露敏感信息如secret_key_base和数据库密码,最终导致远程代码执行。成功利用需要应用使用libvips处理图片上传且接受不受信任的用户上传。

VulnCheck检测到攻击者上传精心构造的图片,利用libvips加载不安全的内容,同时发现即使补丁版本8.1.3.1修复了libvips文件读取,但variation-key的Marshal反序列化漏洞仍存在,RCE gadget在有效签名下仍可执行。

攻击者还利用CVE-2026-5027和CVE-2025-3248等历史漏洞部署Python凭证窃取器、代理和SimpleHelp远程访问工具,并禁用auditd创建取证盲区,随后横向移动以扩大加密货币挖矿网络。

行业启示

AI开发平台如Langflow正成为攻击者的新焦点,因其集成敏感凭证和云环境访问权限,安全团队需优先评估和修补这些平台的漏洞,并实施严格的输入验证和最小权限原则。

针对Web框架的漏洞利用呈现跨平台趋势,Ruby on Rails和Langflow同时受影响,表明攻击者正在系统性挖掘主流技术栈中的高危缺陷,企业应建立持续的漏洞监控和快速响应机制。

攻击者利用蜜罐系统进行侦察和凭证收集,提示组织应加强内部网络的分段和监控,特别是对于暴露在互联网上的AI服务,需部署入侵检测系统和定期安全审计。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Security 安全 Open Source 开源 LLM 大模型 Deployment 部署