SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users

Summary

An attacker successfully exploited a remote code execution vulnerability in SGLang, an open-source inference engine. This exploit allowed the attacker to run code on NEAR AI's inference servers and potentially access confidential data processed in a trusted execution environment.

IFF Assessment

FOE

The exploitation of a remote code execution vulnerability leading to data theft is bad news for defenders, indicating a successful attack against a system.

Severity

9.8 Critical (AI Estimated)

The vulnerability allows for remote code execution (Attack Vector: Network) with significant impact on confidentiality, integrity, and availability (Scope: Changed, Confidentiality: High, Integrity: High, Availability: High), making it critical.

Defender Context

This incident highlights the risk of vulnerabilities in open-source inference engines used in AI applications. Defenders should prioritize patching known vulnerabilities in such software and carefully scrutinize the security posture of their AI infrastructure, especially for components handling confidential data.

Read Full Story →