NEAR AI Confirms Data Exposure After SGLang Vulnerability Exploitation
An attacker gained code execution on a NEAR AI inference server. The company confirmed access to data from approximately 100 users and 18 legacy agents.
NEAR AI confirmed a security incident in which an attacker exploited a vulnerability in SGLang that enabled remote code execution (RCE). According to the company, the attacker gained code execution on one inference server on November 8, 2026. NEAR AI disclosed the incident the following day.
The attacker subsequently accessed data belonging to approximately 100 users stored in an internal system. This included email addresses, names, and, for some accounts, plan and balance information. The company also confirmed access to 18 agents from an older deployment of NEAR AI Agents that was not operated in a TEE environment.
SGLang Vulnerability Affected Inference Server
For the affected agents, chat histories, settings, and logs may have been accessed. This increases the risk especially for users who entered sensitive information into conversations with agents.
In its notice, NEAR AI also stated that the attacker may have been able to observe requests processed on the compromised server during November 8. This included requests in confidential mode. However, the company did not confirm that the attacker actually viewed or exfiltrated those requests from the infrastructure.
The logs also contained API keys belonging to two NEAR AI Cloud customers. The company said it revoked both keys. It contacted the affected users.
Company Revoked Sessions, Credentials, and Tokens
NEAR AI says it fixed the vulnerability on its inference servers. Its response included revoking active sessions and administrator credentials, as well as rotating agent tokens. The company did not disclose technical details of the attack, the affected SGLang versions, or a CVE identifier.
NEAR AI found no indications of unauthorized access to stored Private Chat conversations or uploaded files. However, this statement alone is not proof that these data were completely unaffected.
No Public Advisory from the SGLang Project
The SGLang project repository does not yet contain a security notice directly related to this incident. Specific configurations and versions affected by the vulnerability in SGLang, as well as technical mitigations recommended by the project itself, are therefore not publicly known.
SGLang operators should monitor for any CVE or project advisory and verify their deployment configurations. Any clarification of the incident’s scope from NEAR AI will also be important, including whether the attacker actually accessed requests processed in confidential mode.
The incident also illustrates the limits of an isolated TEE environment: if a vulnerability affects software running directly inside the enclave, using a TEE alone may not prevent data from being processed by compromised code.
Sources
- NEAR AI — Security Incident Report: SGLang Vulnerability – Primary confirmation of the incident, the scope of affected accounts and agents, possible observation of requests, key revocation, and remediation steps taken.
- SGLang GitHub Security Advisories – As of November 11, 2026, no publicly published security advisory from the SGLang project is listed.
- The Hacker News – Independent secondary coverage of NEAR AI’s announcement and the incident’s context.
Verified and updated: November 11, 2026 06:30
