SGLang RCE Exploit Breaches NEAR AI, Hits 18 AI Agents

The breach happened last Sunday. NEAR AI said the attacker could also have seen requests processed on that server that day, including confidential-mode requests handled inside a trusted execution environment. The company noted the TEE does not protect against a flaw in software running inside it, and SGLang runs inside it.

What the attacker reached

• An internal system supporting Private Chat, a product NEAR AI had already retired. About 100 users' names and emails, and for some their plan and balance, were downloaded. There is no sign conversations or files were accessed.
• 18 AI agents on an older NEAR AI Agents deployment outside the TEE, yielding chat history, settings and logs.
• Two customers' API keys exposed in logs, both since revoked.

The company's response

NEAR AI revoked all sessions and admin credentials, replaced agent tokens, took Private Chat offline and deployed an SGLang fix on all inference servers. It reported the flaw to the SGLang project but has not given a CVE or affected versions.

What to do

1. If you run SGLang, watch the project's advisories and never expose it directly to the internet.
2. Do not rely on a TEE alone; the software inside it must be secure too.
3. Keep API keys out of logs and scan your logs for exposed secrets.
4. NEAR AI users should be wary of unexpected messages about their account or balance.