NEAR AI has confirmed a security breach in which an attacker exploited a flaw in an open-source AI inference engine to break into its servers, downloading account details for around 100 users and data from 18 AI agents.
The company, the AI arm of the NEAR blockchain ecosystem, published a security incident report on November 9. It said the attacker exploited a remote code execution vulnerability in SGLang, the open-source engine NEAR AI uses to serve large language models, gaining code execution on a single inference server on November 8.
From there, the intruder reached an internal system behind NEAR AI's previously deprecated Private Chat feature and downloaded account details — email addresses, names, and for some accounts plan and balance information — for approximately 100 users.
What the attacker reached
The breach also touched 18 agents on a legacy deployment of NEAR AI Agents that was not running inside the company's confidential computing environment. Data taken from those agents included chat histories, settings and logs.
Two NEAR AI Cloud customers had API keys exposed in logs. NEAR AI said both keys have since been revoked.
Perhaps the most striking detail is what the attacker could see on the compromised server. On November 8, the intruder could have observed user requests being processed on that server — including requests running in confidential mode, the privacy feature NEAR AI markets as keeping prompts hidden even from the operator.
The confidential computing caveat
NEAR AI said it found no indication that stored Private Chat conversations or uploaded files were accessed. But the incident underlines a limit of confidential computing: hardware enclaves protect data from the host and operator, not from a vulnerability in the software running inside the protected environment itself.
The company said it has reported the SGLang vulnerability to the SGLang project. It has not published a CVE number for the flaw, so the exact bug remains unconfirmed. In May 2026, Carnegie Mellon's CERT Coordination Center published an advisory on three SGLang vulnerabilities, including two remote code execution flaws, though NEAR AI has not said whether its incident involved any of those.
Contained, patched and restored
After detecting the intrusion, NEAR AI revoked all active sessions and administrative credentials, replaced agent tokens, took Private Chat offline and deployed a fix for the SGLang flaw across all inference servers. Its Cloud service, paused during the investigation, has been restored, and affected users have been contacted directly.
NEAR AI said the underlying NEAR Protocol blockchain and other NEAR services were not affected, and there was no established connection to the recent Binance hack.
The incident is a reminder that AI infrastructure is now a target in its own right: the software that serves models can be as consequential a weak point as the models themselves.



