RC RANDOM CHAOS

SGLang RCE Vulnerability Breaches NEAR AI Systems

· via The Hacker News

Original source

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

The Hacker News →

An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, to gain unauthorized access to NEAR AI’s servers. The vulnerability proved particularly dangerous because SGLang runs within trusted execution environments (TEEs) meant to isolate sensitive operations. Yet the security of a TEE cannot protect against flaws in the software running inside it—the attacker accessed requests being processed in confidential mode.

The breach exposed account details for roughly 100 users of NEAR AI’s discontinued Private Chat service, including email addresses and names; some records also contained billing information. The attacker further accessed 18 AI agents running on an older NEAR AI deployment outside the TEE infrastructure, extracting chat histories, configuration data, and system logs. Two API keys embedded in logs were also compromised before being revoked.

NEAR AI responded by immediately revoking sessions, credentials, and all agent tokens, deploying patches to SGLang across its servers, and taking Private Chat offline. However, several critical details remain undisclosed: the SGLang vulnerability lacks a CVE identifier, affected versions have not been specified, and whether the project has released a corresponding patch remains unclear. The incident underscores how infrastructure security mechanisms can be undermined when underlying software contains exploitable flaws.

Read the full article

Continue reading at The Hacker News →

This is an AI-generated summary. Read the original for the full story.