Hacker breaks into NEAR AI through a flaw in an open-source AI engine, reaching account details of about 100 users and data from 18 agents

NEAR AI, the artificial intelligence arm of the NEAR Protocol ecosystem, says a hacker got into its systems on November 8 through a vulnerability in SGLang, an open-source engine its servers use to run AI models. In a report published on Monday, the company said the attacker ran code on one inference server and from there reached an internal system behind Private Chat, a feature it had already deprecated.

About 100 users had their names and email addresses downloaded, and some also had plan and balance information taken. The attacker also reached 18 agents on a legacy NEAR AI Agents deployment that did not run inside the company's confidential computing environment, exposing their chat history, settings and logs. Two NEAR AI Cloud customers had API keys exposed in logs; those keys have been revoked. According to the company, requests handled by that server on November 8, including confidential-mode ones, may also have been visible to the attacker.

NEAR AI says it closed the access, patched the flaw on all inference servers, rotated credentials, took Private Chat offline and has contacted affected users. Its confidential computing, it acknowledged, "does not protect against a vulnerability in the software running inside the environment."