VYPR
researchPublished Aug 28, 2026· 1 source

Two Critical Flaws Grant Root RCE on Unitree Humanoid Robots, One Exploitable via Bluetooth

Security researcher Olivier Laflamme has disclosed two independent vulnerabilities, CVE-2026-76639 and CVE-2026-76640, that allow root remote code execution on Unitree G1 EDU humanoid robots, with one flaw being exploitable over Bluetooth Low Energy.

Security researcher Olivier Laflamme has uncovered two critical vulnerabilities in the Unitree G1 EDU humanoid robot, both of which allow for root remote code execution (RCE). The first flaw, tracked as CVE-2026-76639, enables network-adjacent RCE through a path-traversal vulnerability in the chat_go utility, which leads to code execution within the bashrunner component. This chain ultimately grants an attacker root privileges on the robot's Locomotion PC.

The second vulnerability, CVE-2026-76640, presents an even more concerning attack vector as it can be exploited over Bluetooth Low Energy (BLE). This flaw bypasses standard Bluetooth pairing requirements, allowing an attacker in proximity to initiate a bootstrap interaction. While the initial bootstrap material is protected, the subsequent Wi-Fi provisioning operations require an authenticated BLE state, which the attacker can achieve.

Laflamme's research indicates that a buffer overflow in the Wi-Fi provisioning code, once authenticated via the BLE chain, results in root execution on the Locomotion PC. This BLE-based RCE is particularly alarming due to the potential for widespread exploitation without requiring prior network access or physical tampering.

Compounding the issue, an authorization gap was identified in Unitree's cloud service. During Laflamme's research, the cloud service would accept a valid Unitree account for a key-recovery request without verifying if that account actually owned the target robot. This allowed an attacker to recover key material associated with a different G1 EDU robot, which could then be used to establish the necessary authenticated BLE state for the exploit chain.

While Laflamme noted that Unitree patched this specific cloud authorization flaw in July 2026, the disclosure of the two RCE vulnerabilities on August 27, 2026, leaves G1 EDU owners in a precarious position. As of the disclosure date, a confirmed fixed firmware release addressing both CVE-2026-76639 and CVE-2026-76640 has not been publicly verified or communicated by Unitree.

Unitree's official product pages differentiate between the G1 and G1 EDU models, and the broader applicability of these vulnerabilities to other Unitree robot models remains unconfirmed. The Hacker News has reached out to Unitree for comment on the affected product scope, the status of remediation, and the specific firmware versions that address these critical security flaws.

This discovery highlights the growing security risks associated with increasingly complex robotic systems, particularly those with network connectivity and Bluetooth capabilities. The ability to gain root access on such devices could lead to significant operational disruption, data theft, or even physical harm if the robots are deployed in critical environments.

Synthesized by Vypr AI