NEAR AI has restored access to its Cloud service following a security incident that began on November 8, 2026. An attacker exploited a remote code execution vulnerability in SGLang, the open-source inference engine running on the company’s servers, and gained a foothold on one inference server before moving into connected internal systems.

From there the intruder reached infrastructure linked to the previously discontinued Private Chat feature and downloaded account information belonging to approximately 100 users. The records included names and email addresses, and in some cases subscription plan and account balance details. Investigators found no evidence that stored Private Chat conversations or uploaded files were accessed.

Legacy Agents and API Keys Compromised

The attacker also reached 18 agents running on an older NEAR AI Agents deployment that sat outside the company’s confidential computing environment. Conversation histories, configuration settings and operational logs from those agents were retrieved. Logs associated with the incident further exposed API keys belonging to two Cloud customers; both credentials have been revoked.

On the day of the intrusion the attacker could have observed requests being processed on the compromised server, including those running in confidential mode. NEAR AI distinguished this potential live-view exposure from any access to data already stored on its systems.

The company noted that trusted execution environments are designed to shield information from infrastructure operators and hosts, not from flaws inside the software executing within those environments. The SGLang vulnerability allowed code to run inside one inference server and therefore to see requests in processing.

Containment, Patching and Service Restoration

NEAR AI blocked the attacker’s access, applied a fix for the SGLang vulnerability across all inference servers, and replaced affected credentials. Active sessions and administrative credentials were revoked, agent tokens were rotated, and the Private Chat service was taken offline. A full forensic review was completed along with the remaining credential rotations.

Cloud operations, which had been paused while the investigation and remediation work continued, have been restored. The company contacted affected users and the owners of the 18 agents individually. In a public update it stated that anyone who has not received direct notification needs to take no action. Instructions for those contacted were issued from the firm’s security email address. Regulatory notifications will be made where required, and users were advised to treat unsolicited messages that reference account details or balances with caution.

NEAR Protocol and other NEAR services outside the AI infrastructure were not affected. The vulnerability has been reported to the SGLang project maintainers. Going forward, NEAR AI said its security analysis will treat inference software, surrounding services, dependencies and deployment configurations with the same level of scrutiny applied to the enclave itself. The code, dependencies and configurations for its confidential inference stack remain publicly available for verification.