The security incident that forced the temporary suspension of NEAR AI’s cloud inference service led to the theft of information from 18 AI agents and approximately 100 user accounts, the company disclosed. NEAR AI also noted that confidential AI requests processed on one of its servers may have been visible to the attacker during the breach.
The company said an attacker on Nov 8 exploited a remote code execution vulnerability in SGLang, an open source inference engine used by the firm. To resolve the issue, NEAR AI patched the flaw across its servers, rotated compromised credentials, and restored the cloud service. The breach was limited to internal infrastructure and did not impact the NEAR Protocol.
