SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users
An attacker exploited a remote code execution vulnerability in the SGLang open-source inference engine, compromising NEAR AI's infrastructure and potentially accessing confidential data.
A sophisticated attack has targeted NEAR AI's infrastructure, leveraging a remote code execution (RCE) vulnerability within the SGLang open-source inference engine. The breach, which occurred last Sunday, allowed an attacker to execute arbitrary code on one of NEAR AI's inference servers.
This exploitation raises significant concerns, as the compromised server was processing user requests, including those submitted in a confidential mode. This mode is designed to process sensitive information within a trusted execution environment (TEE), intended to provide a high level of security and privacy. The attacker's ability to access data within this environment represents a severe security failure.
NEAR AI confirmed that the incident affected approximately 18 agents and around 100 users. The full scope of the data exfiltrated is still under investigation, but the potential exposure of confidential requests processed in a TEE is a critical aspect of the breach.
SGLang is an open-source project designed to simplify the deployment and management of large language models (LLMs). Its inference engine is used by various organizations to run AI models efficiently. The vulnerability exploited in this incident highlights the inherent risks associated with deploying complex AI infrastructure, especially when third-party open-source components are involved.
The incident underscores the growing threat landscape surrounding AI infrastructure. As organizations increasingly rely on LLMs and related technologies, securing these systems becomes paramount. Attackers are actively seeking vulnerabilities in the software supply chain and inference engines that power these advanced AI capabilities.
While NEAR AI has not yet released specific details on the vulnerability in SGLang or the timeline for a patch, the company is investigating the incident. Users and organizations utilizing SGLang or similar inference engines are advised to remain vigilant and monitor for any security advisories or updates from the SGLang project and NEAR AI.
This breach serves as a stark reminder of the security challenges posed by the rapid adoption of AI technologies. The potential for attackers to gain access to sensitive data processed within TEEs, even when designed for confidentiality, necessitates a thorough review of security protocols and the implementation of robust defenses for AI deployments.