NEAR AI Reports SGLang RCE Breach Exposing In-Flight Requests
NEAR AI says an attacker exploited a remote code execution vulnerability in SGLang, accessing account details of about 100 users, 18 legacy agents, and in-flight requests, including confidential-mode traffic. No stored conversations or uploaded files were accessed, the company said.
NEAR AI published a security incident report on November 9 describing an attack from the day before: an attacker exploited a remote code execution flaw in SGLang, an open-source inference engine, and gained code execution on one inference server. The attacker downloaded account details — emails, names, and partial plan and balance data — for some 100 users of the retired Private Chat feature. The attacker also accessed 18 agents in a legacy deployment outside the confidential computing environment, exposing chat history, settings, logs, and two customer API keys, since revoked.
On November 8, the company said, the attacker could observe requests being processed, including those in confidential mode. NEAR AI stated there was no indication that stored conversations or uploaded files were accessed. "Confidential computing protects data from the operator and the host. It does not protect against a vulnerability in the software running inside the environment," the report said.
As of November 10, no CVE identifier, affected version list, or public patch from the SGLang project had been published; the latest release, v0.5.23, dates from November 2. SGLang has a prior record of such flaws: CVE-2026-3059 in its ZMQ broker and CVE-2026-3060, both patched earlier this year.