NEAR AI says unauthorized access hit internal systems, pauses cloud inference

NEAR AI said it identified unauthorized access to its internal systems and temporarily paused NEAR AI Cloud inference as a precaution, according to an official incident report and the project’s social update. The company said the issue did not affect NEAR Protocol.

In its public disclosure, NEAR AI said the incident was tied to an SGLang vulnerability and that customer prompts were protected inside trusted execution environments, or TEEs, while the company investigated. The initial update framed the pause as a containment step rather than a broader protocol interruption.

Cloud inference was taken offline while the investigation continued

The company said NEAR AI Cloud inference was paused after the unauthorized access was detected, with the service later restored after the issue was addressed. NEAR AI also said it would provide another update within two hours, indicating the situation was being handled as an active security incident rather than a routine maintenance event.

NEAR AI’s report distinguishes between the cloud inference service and the wider NEAR ecosystem, saying the protocol itself was unaffected. That matters for readers trying to separate an application-layer security event from a network-level disruption.

The disclosure adds to a growing number of security incidents in AI infrastructure, where access control, inference availability and confidential-computing claims can all become part of the same operational risk picture. In this case, NEAR AI’s public statement centered on internal-system access, a temporary service pause and the restoration of cloud inference after remediation.