Observed incident

NEAR AI reports account data and legacy agent histories accessed through an inference server

Reported

Published here

What happened

NEAR AI says an intruder exploited a flaw in its SGLang inference software and reached internal systems. The intruder downloaded account details for about 100 users of its retired Private Chat service and retrieved histories, settings and logs for 18 agents on a legacy deployment.

Parties and reporting

Where this happened

Platforms and supply chain. A provider, package, marketplace or repository was compromised. Affected customers are wherever its users are; the sources do not break them down by country. Explore the world map.

AI tools and systems: SGLang inference engine, NEAR AI Cloud, Legacy NEAR AI Agents and Private Chat

Outcome

NEAR AI deployed a fix across its inference servers, replaced agent tokens and credentials, revoked two customers’ API keys that appeared in logs, and notified affected users. It reported that cloud access was restored.

Operating lesson

Limit what an inference server can reach, keep credentials out of logs, and include retired services in access reviews.

Evidence and limits

The 18 agents ran outside confidential computing. NEAR AI says an intruder could have observed November 8 requests, including confidential ones, on the affected server. It found no indication of access to stored Private Chat conversations or uploaded files. Its report supplies no CVE or upstream fixed version.

Sources