Last Sunday, an attacker exploited a remote code execution vulnerability in SGLang, an open-source inference engine, to gain unauthorized access to a NEAR AI inference server. This breach allowed the attacker to execute code and potentially view sensitive requests being processed on the server, including those made in a confidential mode designed to enhance privacy through a trusted execution environment. As a result, data from 18 agents and around 100 users was compromised. This incident raises serious concerns about the security of open-source software and the potential risks involved in handling sensitive data. Companies using SGLang or similar technologies need to assess their security measures to prevent future breaches.