NEAR AI has restored its Cloud service after containing an unauthorized access incident that exposed information belonging to approximately 100 users and data from 18 AI agents, according to a November 9 security report published by the organization.
The breach occurred on November 8, when an attacker exploited a remote code execution vulnerability in SGLang, an open-source inference engine operating on NEAR AI servers. The flaw allowed the unauthorized party to execute code on an inference server and gain access to internal infrastructure. The attacker subsequently reached an internal system tied to NEAR AI’s discontinued Private Chat service, downloading account details including names, email addresses, and select subscription and balance details for roughly 100 users.
Additionally, the attacker accessed 18 agents operating on a legacy deployment outside the company’s confidential computing or Trusted Execution Environment. Retrieved data for these agents included chat histories, configuration settings, and operational logs, which exposed API keys for two NEAR AI Cloud customers. The investigation also confirmed that requests being processed on the compromised server on November 8, including some running in confidential mode, could have been observed. No evidence indicated that stored Private Chat conversations or uploaded files were accessed.
In response, NEAR AI suspended Cloud inference operations, revoked active sessions and administrative credentials, replaced agent tokens, and invalidated the exposed API keys. Engineers deployed a security patch for the SGLang vulnerability across all inference servers, completed credential rotations, and reported the flaw to the SGLang project.
The company clarified that the incident was isolated to NEAR AI and did not affect the underlying NEAR Protocol blockchain or other ecosystem services. Affected users and agent owners have been notified directly with security guidance.
