Near AI said an attacker exploited a vulnerability in SGLang to run code on an inference server and reach internal systems. The attacker downloaded account information for about 100 users of its discontinued Private Chat service and accessed 18 legacy AI agents outside its trusted execution environment. Logs from those agents exposed API keys for two Cloud customers; Near AI said both keys have been revoked.
Near AI found no indication that stored Private Chat conversations or uploaded files were accessed. It said requests processed on the compromised server on Nov. 8, including confidential-mode requests, could have been observed, but did not say how many. The company said Cloud access was restored after it patched its servers, and acknowledged that confidential computing does not protect against flaws in software running inside the protected environment.
