NEAR AI disclosed unauthorized access to its internal systems on November 9 and paused its cloud inference service as a precaution while it investigates.
The company announced the incident on X, stating it had "identified unauthorized access to NEAR AI internal systems" and that inference on NEAR AI Cloud would remain offline during the investigation. NEAR AI promised an update within two hours of the initial disclosure, posted at 11:14 PM UTC.
According to the company, customer prompts were not exposed. NEAR AI Cloud routes inference requests through hardware-isolated Trusted Execution Environments, or TEEs, which the company describes as sealed enclaves that prevent access by the cloud host, the GPU operator, and NEAR AI itself.
What NEAR AI Cloud Does
NEAR AI sells private inference. Where a typical cloud AI provider asks you to trust that it will not read your prompts, NEAR AI runs models inside Intel TDX confidential virtual machines and NVIDIA GPUs in confidential-computing mode. The company delivers a cryptographic attestation with every response, letting clients verify that their data was processed inside a verified enclave.
The platform offers 51 models, 15 of which run in TEEs. Customers include Brave, which began routing its Leo AI assistant through NEAR AI's confidential pipeline in November 2025 and expanded the integration to all production users selecting the GLM 5.3 Flash model last month. Corbits, an enterprise collaboration platform, also integrates NEAR AI private inference for its AI agents.
"Brave's and NEAR's values are aligned on the importance not just of privacy, but also verifiability for all," the company wrote in a blog post announcing the Brave partnership last month.
The Outage
NEAR AI's status page listed every major service component in a major outage state as of 10:50 PM UTC on November 9: Cloud API, Chat API, all private inference models including DeepSeek-V4.1-Flash, multiple Qwen variants, and GLM 5.3 Flash, plus the attestation service that signs hardware proofs. Only the Agent Servers remained operational.
Brave's own status page showed Leo AI running at 99.994% uptime at the time of the NEAR AI disclosure, suggesting the outage may not have immediately affected all Brave Leo users. Brave routes Leo traffic through its own AI Gateway before it reaches NEAR AI, and the browser supports multiple model backends.
The Distinction NEAR AI Is Drawing
The company's statement separates "internal systems" from the confidential inference pipeline. That distinction matters. If the attackers accessed corporate infrastructure but not the enclaves that process customer prompts, the TEE architecture did what it was designed to do: limit the blast radius of a breach.
But that claim cannot be independently verified until NEAR AI publishes more detail. The company has not said what systems were accessed, how the attackers got in, or whether any customer-facing data was involved.
NEAR AI also emphasized that NEAR Protocol, the layer-1 blockchain the company is affiliated with, was not affected. NEAR Protocol and NEAR AI share branding and some history, but the blockchain and the inference platform run on separate infrastructure.
Context
AI inference platforms have become high-value targets. In July, Hugging Face disclosed an intrusion driven end-to-end by an autonomous AI agent that exploited code-execution paths in its dataset processing pipeline. Security researchers have documented 76 AI-agent-related security cases as of late October 2026, including 13 confirmed real-world incidents, according to one tracker.
NEAR AI's pitch to customers rests on verifiability: the idea that you do not have to trust the company's promises because you can check the attestation. That model faces its first real test now. If the breach truly stopped at corporate systems and never touched the enclaves, the architecture held. If the investigation reveals something more, the value proposition looks different.
The company said it would provide an update within two hours. As of publication, no follow-up had appeared.
