SGLang flaw exploited to access NEAR AI infrastructure

SGLang access reached accounts, agents and live requests

NEAR AI said SGLang was the open-source inference engine running on its servers. The vulnerability gave the unauthorized party code execution on one inference server and access from there to NEAR AI’s infrastructure.

The attacker reached an internal system behind the previously deprecated Private Chat service and downloaded email addresses, names, and, for some accounts, plan and balance information for approximately 100 users.

Data retrieved from the 18 agents on the legacy deployment included chat history, settings and logs. Logs exposed NEAR AI Cloud API keys belonging to two customers, and NEAR AI revoked those keys.

NEAR AI took Private Chat offline and fixed SGLang

After detecting the access, NEAR AI revoked all active sessions and administrative credentials, replaced all agent tokens, took Private Chat offline, and deployed a fix for the SGLang vulnerability across all inference servers.

The company said it completed rotation of all remaining credentials and a full forensic review. It reported the vulnerability to the SGLang project and restored access to NEAR AI Cloud.

The Hacker News said NEAR AI has not identified the vulnerability by CVE number or named the affected SGLang versions. Carnegie Mellon University’s CERT Coordination Center documented six SGLang vulnerabilities in July, including bugs that could let attackers execute commands or expose credentials depending on server configuration. Decrypt said NEAR AI did not identify which flaw was used here.

Live confidential requests may have been visible on Nov 8

The Hacker News said NEAR AI has not said whether the affected server handled chats from Brave’s Leo assistant. Brave said Leo’s end-to-end encrypted AI chat was available in its Release, Beta and Nightly channels, while its GLM 5.3 Flash model ran inside a sealed, Nvidia-backed hardware enclave operated by NEAR AI.

NEAR AI said it has notified users whose account details or agents it confirmed were accessed and will make any regulatory notifications that are required. It also contacted the owners of all 18 affected agents individually.

NEAR AI did not say whether the attacker used the two exposed Cloud API keys. The Hacker News said NEAR AI has not said whether people whose requests passed through the affected server were among those contacted or whether the server handled Brave Leo chats.