AI Stack Alert High For CISO, SOC / IR, AppSec, AI Eng Unconfirmed AI-assisted
SGLang RCE Exploited to Breach NEAR AI Inference Server, Exposing Agent and User Data
An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and ran code on one of NEAR AI's inference servers. The attacker could also have viewed requests processed on that server that day, including requests sent in confidential mode. The source reports data theft affecting 18 agents and about 100 users.
Why it matters
Organizations running SGLang-based inference services may be exposed to code execution on model hosts. Requests processed there, including those meant to run in confidential mode, may be visible to an attacker.
Do this
- Identify all SGLang deployments in your environment and confirm their versions against the vendor's fixed release.
- Patch or upgrade SGLang on inference servers as soon as a fixed version is available, and restrict network access to management interfaces.
- Review logs on inference hosts for unexpected process execution or outbound connections during the period of the incident.
- Rotate credentials and API keys that were handled by affected inference servers, and review any data sent through them.
- Do not rely on confidential-mode or TEE processing alone as protection when the host runtime is compromised; assess the full request path.
Signals
- NoCISA KEV
- 33/100Priority
- 60%Confidence
Evidence
“An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and ran code on one of NEAR AI's inference servers last Sunday.”
“The attacker could also have seen requests being processed on that server that day, including requests sent in confidential mode, the company said earlier this week.”
Sources
Affected / related
SGLang, NEAR AI
Published 2026-11-10 19:19 UTC · updated 2026-11-12 17:35 UTC · AI-assisted summary, checked against quoted sources · Unconfirmed: single non-official source · How we score