Disclosed weakness · no CVE recorded
NEAR AI reports exploitation of a code-execution flaw in SGLang
NEAR AI (reporting provider); SGLang project · SGLang inference engine as deployed by NEAR AI
First documented disclosure · Review checkpoint
Summary
NEAR AI traced unauthorized access on one inference server to the inference engine running inside its environment.
Severity and exploitation evidence
No CVSS score recorded in this review.
Reported activity: NEAR AI documented actual exploitation on one inference server. The report does not establish exploitation of other deployments.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
A flaw let a remote party run code within the inference server. The provider did not disclose the technical mechanism.
Conditions required
Access to a susceptible deployment was required. The public report does not identify the interface or configuration.
Potential impact
The provider confirmed access to account data and legacy agents. It described possible observation of live requests; it did not establish access to stored Private Chat conversations.
Affected versions and configurations
The affected upstream version and broader range are unknown.
Fix and mitigation status
NEAR AI reports fixing its own fleet. No upstream fixed version is identified.
Practical next steps
For the affected service, follow the provider’s response. Other operators should review inference-server exposure, accessible credentials and upstream guidance for their own build.
Operating lesson
Treat software running inside a confidential-computing environment as part of the security boundary.
Evidence and limits
The provider reported the issue upstream. No CVE, upstream release or relationship to another SGLang advisory is established.
Related incident and research records
Related records can include laboratory demonstrations; a relationship does not itself establish exploitation.
Sources
- NEAR AI incident report
Primary source ·