SGLang Engine Flaw Exposes Near AI User Data
Breach compromised information from 100 accounts and 18 AI agents following a vulnerability exploit.
Near AI confirmed a security flaw that allowed unauthorized access to sensitive data within its systems. An attacker used a remote code execution vulnerability in the SGLang engine to compromise the company's infrastructure.
The attack resulted in the exposure of names and email addresses for approximately 100 users. Additionally, the attacker obtained information regarding subscription plans and balances for accounts linked to the platform.
Eighteen artificial intelligence agents had their conversation histories, configurations, and logs accessed. These agents were running on an older implementation that lacked the company's confidential computing protocols.
Cloud API keys belonging to two clients were also exposed during the incident. Near AI immediately revoked these credentials to prevent further unauthorized access to cloud services.
The company admitted that hardware isolation does not prevent flaws in software running within a protected environment. The case highlights that vulnerabilities in model processing engines can bypass privacy protections focused on encryption.
Spotted an error? Write to suporte@criptoico.com.br.
