NEAR AI says SGLang flaw led to access to internal systems

NEAR AI said the attacker exploited a remote-code-execution vulnerability in SGLang, the open-source inference engine running on its servers.[1] The flaw enabled code execution on one inference server and access to NEAR AI’s infrastructure from there, according to the company.[1] NEAR AI said the attacker could have observed requests processed on the accessed server on November 8, including requests in confidential mode.[1] The company explained that confidential computing does not protect against vulnerabilities in software running inside the protected environment.[1]

NEAR AI said account details for approximately 100 users were downloaded, including names and email addresses, as well as plan and balance information for some accounts.[1] The attacker also reached an internal system associated with the previously deprecated Private Chat feature.[1] NEAR AI said it found no indication that stored Private Chat conversations or uploaded files were accessed.[1] Separately, data retrieved from 18 agents on a legacy NEAR AI Agents deployment included chat histories, settings and logs.[1] That deployment was not running in NEAR AI’s confidential-computing environment.[1]

Two NEAR AI Cloud customers had API keys exposed in logs, and NEAR AI said it revoked those keys.[1] After detecting the breach, the company revoked active sessions and administrative credentials, replaced agent tokens and took Private Chat offline.[1] It deployed a fix for the SGLang vulnerability across all inference servers, completed rotation of its remaining credentials and said it finished a full forensic review.[1] NEAR AI said it contacted the owners of the 18 agents individually and notified users whose account details or agents it had confirmed were accessed.[1]

In an initial update at 04:44 a.m. IST, NEAR AI said it had identified unauthorized access and paused cloud inference as a precaution while investigating.[2] At 06:50 a.m. IST, it said the unauthorized access had been closed but the investigation was still being completed.[2] At 09:02 a.m. IST, it said its cloud service was back online.[2] NEAR AI said it would make any required regulatory notifications and reported the vulnerability to the SGLang project.[1] The company said its confidential inference stack’s code, dependencies and deployment configurations remain public for verification.[1]