Attacker Exploits SGLang Flaw to Breach NEAR AI, Exposing Data from 18 AI Agents and About 100 Users
NEAR AI has disclosed that an attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and ran code on one of its inference servers last Sunday. The company says the attacker could also have seen requests processed on that server that day, including those sent in confidential mode. That mode uses a trusted execution environment (TEE) to keep requests private, but the company noted a TEE does not protect against a vulnerability in the software running inside it.
The attacker then moved further in. They reached an internal system behind Private Chat, a product the company had already deprecated, and downloaded account details for about 100 users. These were email addresses and names, and for some accounts plan and balance information. The company found no sign that stored conversations or uploaded files were accessed. The attacker also reached 18 AI agents on an older NEAR AI Agents deployment that did not run inside the TEE, taking chat history, settings and logs. Two customers' API keys exposed in logs have been revoked.
NEAR AI revoked all sessions and administrator credentials, replaced agent tokens, took Private Chat offline, deployed a fix on all inference servers and completed a forensic review. Service has been restored. The company has not named a CVE number or affected SGLang versions, nor said whether the SGLang project has released a fix. Affected users have been contacted directly and are asked to be wary of unexpected messages about their account or balance.