INSPECTING ARCHIVED INTELLIGENCE (OUTDATED VERSION).

NEAR AI Breached via SGLang RCE Exploit

| 2026-11-10 20:39 LOW LOW AI-ENABLED ATTACK · AUTONOMOUS DATA BREACH EXPLOITED VULNERABILITY
Executive Summary
AI-generated
An attacker exploited a remote code execution vulnerability in SGLang, an open-source engine that runs AI models and generates responses to user requests, on November 9. The breach exposed account information belonging to about 100 users and data from 18 AI agents after the intruder executed commands on one inference server and then reached other parts of Near AI's infrastructure. The attacker accessed confidential mode requests processed inside a trusted execution environment (TEE), which was not protected against software vulnerabilities, allowing them to reach internal systems behind Private Chat, a deprecated product. The company has confirmed that account details for about 100 users were downloaded, including email addresses and names, as well as data from the 18 AI agents, including chat history and settings.
Technical Mitigations AI-generated
• Monitor for suspicious activity on inference servers running GLM 5.3 Flash model inside a TEE operated by NEAR AI. • Block or hunt for requests being processed in confidential mode that were sent through the affected server last Sunday. • Rotate all agent tokens and take Private Chat offline after detecting an attacker's access to account details and data from 18 AI agents.
Intelligence Metadata
Actors / Malware / CVEs / Campaigns
Target & Sectors
Global Scope
Incident Timeline
‎November 9
An attacker exploited a remote code execution vulnerability in SGLang to target 18 AI agents, exposing user details and data.
tactic Remote Code Execution
organisation SGLang
‎2026/11/10
An attacker exploited a remote code execution flaw in SGLang, an open-source inference engine used by Near AI, to download account details for about 100 users and data from 18 AI agents.
victims 100 users
organisation SGLang
organisation Vulnerability / Cloud Security
organisation Near AI
organisation Private Chat
Tactical Metrics
Metrics
victims
100
Users
Intelligence Sources