SGLang RCE Breach Compromises NEAR AI Systems
🔒 An RCE bug in SGLang, the open-source inference engine, was exploited to run code on a NEAR AI inference server, potentially exposing requests processed that day, including those in confidential mode. The attacker moved laterally to a deprecated Private Chat system and an older NEAR AI Agents deployment, exfiltrating about 100 user account details and data from 18 agents. NEAR AI revoked credentials, deployed a fix, rotated tokens, and completed forensics, and has notified impacted users.
