NEAR AI security incident report: SGLang vulnerability

Evidence record

Read the original source ↗

Claim-level citations (3)

• supportsNEAR AI confirms incident and containment: NEAR AI says a compromised SGLang inference server could expose requests being processed on November 8, including confidential-mode requests, while reporting no evidence of access to stored Private Chat conversations or uploads.

  Security Incident Report: What was affected and Confidential inference sections

• supportsNEAR AI confirms incident and containment: NEAR AI reported SGLang remote code execution on an inference server followed by unauthorized access to internal customer and legacy-agent systems.

  Security Incident Report; What was affected

• supportsNEAR AI confirms incident and containment: The first-party report identifies roughly 100 users' account details, 18 legacy agents outside TEE, and two customers' API keys in logs, with no reported access to historical stored Private Chat content.

  What was affected; What we did next