AI Security AI安全 4h ago Updated 2h ago 更新于 2小时前 46

Two Unitree G1 EDU Humanoid Robot Flaws Enable Root RCE, One Starts Over Bluetooth Unitree G1 EDU人形机器人两处漏洞可导致Root远程代码执行,其中一处可通过蓝牙启动

Security researcher Olivier Laflamme disclosed two independent root remote code execution (RCE) vulnerabilities (CVE-2026-76639 and CVE-2026-76640) affecting the Unitree G1 EDU humanoid robot CVE-2026-76639 exploits a path-traversal flaw in chat_go to reach bashrunner, achieving root RCE on the robot's Locomotion PC via a network-adjacent attack vector CVE-2026-76640 leverages a BLE proximity-based attack chain: an unauthenticated BLE bootstrap interaction, cloud key-recovery abuse, and a buffer Unitree G1 EDU人形机器人存在两个独立的root级远程代码执行漏洞(CVE-2026-76639和CVE-2026-76640),攻击者可获取机器人最高权限 CVE-2026-76639利用chat_go组件的路径遍历缺陷,通过bashrunner实现Locomotion PC的root RCE CVE-2026-76640通过BLE近场攻击链,利用云服务授权验证缺陷获取密钥材料,再通过Wi-Fi配网代码的缓冲区溢出实现root执行 Unitree已于2026年7月修复云服务授权验证问题,但截至8月27日披露时仍未发布确认的固件修复版本 漏洞影响范围目前仅确认针对G1 EDU型号,

68
Hot 热度
65
Quality 质量
60
Impact 影响力

Analysis 深度分析

TL;DR

  • Security researcher Olivier Laflamme disclosed two independent root remote code execution (RCE) vulnerabilities (CVE-2026-76639 and CVE-2026-76640) affecting the Unitree G1 EDU humanoid robot
  • CVE-2026-76639 exploits a path-traversal flaw in chat_go to reach bashrunner, achieving root RCE on the robot's Locomotion PC via a network-adjacent attack vector
  • CVE-2026-76640 leverages a BLE proximity-based attack chain: an unauthenticated BLE bootstrap interaction, cloud key-recovery abuse, and a buffer overflow in Wi-Fi provisioning code to achieve root execution
  • Unitree patched the cloud account-to-robot ownership verification gap in July 2026, but no confirmed fixed firmware release has been published as of the August 27, 2026 disclosure
  • The vulnerabilities are specific to the G1 EDU model; broader applicability to other Unitree robots remains unconfirmed

Why It Matters

This research highlights critical security gaps in humanoid robots that are increasingly deployed in educational, research, and industrial environments, where root-level compromise could lead to physical safety risks, data exfiltration, or device hijacking. The BLE-based attack chain demonstrates how wireless proximity vulnerabilities can bypass traditional network perimeter defenses, a growing concern as IoT and robotics converge. The lack of a verified firmware fix leaves G1 EDU owners in a vulnerable posture, underscoring the importance of responsible disclosure timelines and vendor responsiveness in the robotics sector.

Technical Details

  • CVE-2026-76639 (Network-Adjacent RCE): Exploits a path-traversal vulnerability in the chat_go component, which allows an attacker to reach the bashrunner service. Successful exploitation yields root-level code execution on the Locomotion PC without requiring authentication.
  • CVE-2026-76640 (BLE-Proximity RCE): A multi-stage attack chain beginning with an unauthenticated BLE write path that accepts bootstrap interactions without Bluetooth pairing. The attacker then abuses a cloud service authorization gap (now patched in July 2026) to recover key material for a target robot using any valid Unitree account. The recovered key establishes authenticated BLE state, enabling Wi-Fi provisioning operations where a buffer overflow yields root RCE on the Locomotion PC.
  • Cloud Authorization Flaw: During the research period, Unitree's cloud service accepted a valid account for key-recovery requests without verifying that the account actually owned the supplied robot, allowing cross-account key material extraction.
  • Affected Firmware: The researcher upgraded the test robot to V1.5.2; however, it was not definitively confirmed whether V1.5.1.1 or other firmware versions are also vulnerable.
  • Scope: Vulnerabilities are confirmed on the G1 EDU model. Unitree's official product page distinguishes the G1 and G1 EDU as separate models, and broader applicability remains unconfirmed.

Industry Insight

  • Robotics manufacturers must treat wireless interfaces (BLE, Wi-Fi provisioning) as attack surfaces requiring the same security rigor as network-facing services, especially for devices capable of physical interaction with humans.
  • Cloud-to-device authentication flows in IoT and robotics ecosystems should enforce strict ownership verification; the patched gap demonstrates how missing authorization checks can enable cross-device key recovery and subsequent local compromise.
  • The absence of a verified firmware fix at time of public disclosure poses a significant risk to deployed units—vendors should prioritize rapid patch distribution and clear communication of affected versions to maintain trust in emerging robotic platforms.

TL;DR

  • Unitree G1 EDU人形机器人存在两个独立的root级远程代码执行漏洞(CVE-2026-76639和CVE-2026-76640),攻击者可获取机器人最高权限
  • CVE-2026-76639利用chat_go组件的路径遍历缺陷,通过bashrunner实现Locomotion PC的root RCE
  • CVE-2026-76640通过BLE近场攻击链,利用云服务授权验证缺陷获取密钥材料,再通过Wi-Fi配网代码的缓冲区溢出实现root执行
  • Unitree已于2026年7月修复云服务授权验证问题,但截至8月27日披露时仍未发布确认的固件修复版本
  • 漏洞影响范围目前仅确认针对G1 EDU型号,其他Unitree机器人产品的适用性尚未验证

为什么值得看

本文揭示了人形机器人在物理安全和网络安全交叉领域的关键风险,BLE和Wi-Fi配网链路的漏洞可能使攻击者在物理接近条件下完全控制机器人系统。对于AI硬件制造商和机器人部署企业而言,这凸显了物联网设备安全开发生命周期和供应链安全审计的紧迫性。

技术解析

  • CVE-2026-76639:网络相邻路径攻击,利用chat_go组件中的路径遍历条件,将恶意输入传递至bashrunner组件,最终在Locomotion PC上实现root级别的远程代码执行。该漏洞作为独立RCE链被披露,同时也被复用为BLE攻击链的初始利用原语。

  • CVE-2026-76640:BLE近场攻击链,初始BLE写入路径在无需蓝牙配对的情况下接受引导交互。攻击者利用Unitree云服务在密钥恢复请求中未验证账户与机器人所有权归属的授权缺陷,获取目标机器人的密钥材料,建立经过认证的BLE状态后触发Wi-Fi配网代码中的缓冲区溢出,实现root执行。

  • 云服务授权缺陷:Unitree云服务平台接受有效的Unitree账户进行密钥恢复请求,但未验证该账户是否真正拥有所提供的机器人。此授权间隙允许攻击者恢复与另一台G1 EDU关联的密钥材料,进而建立后续攻击所需的认证BLE状态。该问题已在2026年7月修复。

  • 固件与影响范围:研究人员将测试机器人升级至V1.5.2版本,但未明确确认V1.5.1.1是否受影响。官方产品页面区分了G1和G1 EDU为独立型号,漏洞对其他Unitree机器人的适用性尚未确认。截至披露日,未找到确认的修复固件版本。

行业启示

  • 人形机器人安全框架亟待建立:随着人形机器人从实验室走向商业和消费场景,物理可接触性(BLE近场)与网络攻击面的结合放大了风险。行业需要建立针对具身智能设备的专项安全标准,涵盖无线通信协议、固件更新机制和权限隔离设计。

  • IoT设备安全开发生命周期(SDL)需强化:路径遍历、缓冲区溢出等经典漏洞出现在2026年的商业机器人产品中,反映出硬件厂商在安全编码实践和代码审查方面的不足。制造商应将渗透测试和漏洞赏金计划纳入产品发布的强制环节。

  • 云服务与边缘设备的信任边界需重新审视:云授权验证缺陷表明,云端身份验证与边缘设备控制之间的信任传递存在断裂。机器人生态系统应采用零信任架构,确保云端操作与设备端执行之间的双向验证和最小权限原则。

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 Robotics 机器人 Research 科学研究