NEAR AI, the AI product and infrastructure arm of the NEAR Protocol blockchain ecosystem, reported a breach on November 9. The company says an attacker exploited a remote code execution flaw, a bug that lets outsiders run their own code, in SGLang, an open-source engine for serving AI models. From one inference server, the attacker reached a deprecated chat service and downloaded names, emails, and in some cases plan and balance details for about 100 users. Chat histories, settings, and logs of 18 legacy agents were also accessed. Two cloud customers' API keys appeared in logs.
NEAR AI says confidential computing, hardware isolation that shields data from the host and operator, does not protect against flaws in the software running inside it. Other teams serving models with SGLang face the same class of risk. The company says it patched all servers, rotated credentials, revoked the exposed keys, took Private Chat offline, and contacted affected users.
Source · theblock.co ↗
