NEAR AI’s November 9 incident report describes a breach that began in a component many users would never see: SGLang, the open-source inference engine running on its servers. An attacker exploited a remote-code-execution vulnerability, gained access to one inference server and then reached other internal systems. NEAR AI says it has closed the access, fixed the vulnerability across inference servers, rotated affected credentials and restored its Cloud service.
The confirmed exposure is narrower than a compromise of the NEAR blockchain but serious in its own right. The company says approximately 100 users had account details downloaded, including email addresses and names and, for some, plan and balance information. The attacker also reached 18 agents on a legacy deployment outside its trusted execution environment, retrieving chat histories, settings and logs. Two Cloud customers’ API keys were exposed in logs and have since been revoked.
This is not evidence that the underlying cryptographic isolation mechanism was defeated. It is evidence that software inside the protected execution path still needs ordinary vulnerability management.
NEAR AI reports no indication that stored Private Chat conversations or uploaded files were accessed. It says affected users were notified, sessions and administrative credentials were revoked and a forensic review completed. The Block independently reported the disclosure November 10. No verified customer-fund theft or NEAR Protocol consensus compromise is established by these reports.
TOKEN RECON ASSESSMENT
The actionable security lesson is not that confidential computing is useless. It is that the inference engine, legacy services and logging system remain part of the trust boundary.
Watch for an upstream SGLang advisory, confirmation of the affected version range, follow-on misuse of exposed details and evidence of the remediation’s effectiveness. Do not describe this as a NEAR chain exploit.
Sources
