Odaily Planet Daily reports: In its latest security incident report, NEAR AI stated that the attacker exploited a remote code execution vulnerability in the open-source inference engine SGLang running on its server, executing code on an inference server and gaining access to related infrastructure. The attacker downloaded account information for approximately 100 users, including email addresses, names, and some account subscription and balance details. Additionally, 18 NEAR AI Agents deployed on outdated infrastructure and not running in a Trusted Execution Environment (TEE) were accessed, with their chat logs, settings, and logs exfiltrated. API keys for two Cloud customers appeared in the logs, and the affected keys have since been revoked.
NEAR AI stated that on November 8, the attacker may have observed user requests being processed by the affected servers, including private mode requests; no stored Private Chat conversations or uploaded files have been found to have been accessed. NEAR AI has contained the unauthorized access, restored NEAR AI Cloud services, patched the vulnerability, and rotated compromised credentials, while also taking down Private Chat. All affected users and owners of the 18 related Agents have been notified. NEAR AI has reported the vulnerability to the SGLang project and will make required regulatory disclosures. (The Block)



