The Ledger
SGLang flaw could have exposed requests in confidential inference, NEAR AI says
NEAR AI says an attacker exploited an SGLang remote code execution flaw to access one inference server and other parts of its infrastructure. The company reports that requests on the server, including confidential-mode requests, could have been observed, while account details for about 100 users and data from 18 legacy agents were accessed.

NEAR AI says a remote code execution flaw in SGLang let an unauthorized party run code on one inference server and reach other parts of its infrastructure. The company says the access was contained, affected credentials were replaced, the vulnerability was fixed and access to NEAR AI Cloud was restored.
In a security incident report published November 9, 2026, NEAR AI said its investigation confirmed that the party exploited the flaw on November 8. On that date, the attacker could also have observed user requests being processed on the accessed server, including requests running in confidential mode. The report describes this as possible observation, not confirmed access to every request.
Account and agent data
The party reached an internal system behind the previously deprecated Private Chat service and downloaded account details for approximately 100 users. The details included email addresses and names, as well as plan and balance information for some accounts.
The report also describes access to 18 agents on a legacy NEAR AI Agents deployment that was not running in NEAR AI's confidential computing, or TEE, environment. For those agents, the data retrieved included chat history, settings and logs. NEAR AI said it found no indication that stored Private Chat conversations or uploaded files were accessed.
Logs exposed NEAR AI Cloud API keys belonging to two customers. NEAR AI said it revoked those keys. The report distinguishes these confirmed data accesses from the possibility that requests in processing on the accessed inference server could have been observed.
Response and notifications
After detecting the incident, NEAR AI revoked all active sessions and administrative credentials, replaced all agent tokens and took Private Chat offline. It deployed a fix for the SGLang vulnerability across all inference servers, completed rotation of all remaining credentials and a full forensic review, and notified users whose account details or agents it had confirmed were accessed.
The company said it will make any regulatory notifications that are required. Affected users were contacted directly from its security email address, security@near.ai, and the 18 agent owners were contacted individually with steps that apply to them. The report advises users to treat unsolicited messages referring to their account or balance with caution.
Limits of confidential computing
Confidential computing protects data from the operator and host, but it does not protect against a vulnerability in software running inside the environment, NEAR AI said. In this case, the SGLang flaw let the attacker run code inside one inference server and access requests being processed there, including confidential ones.
NEAR AI said it reported the vulnerability to the SGLang project. It is now applying the same rigor in security analysis to inference software and services around the enclave as to the enclave itself. The code, dependencies and deployment configurations for its confidential inference stack remain public so that others can verify what is running.
Sources
- Original source · Blog post NEAR AI near.ai