NEAR AI Restores Cloud Inference Hours After an Internal Breach

NEAR AI took its cloud inference service offline late on Monday after finding that someone had reached its internal systems without permission. Three posts on X carried the whole episode. The first said the access had been identified and NEAR AI Cloud paused as a precaution. The second said the access was closed while the investigation ran on. The third said the service was back, the vulnerability fixed across the company's servers and the affected credentials replaced.

Two filings, two dates

The timing depends on which outlet is read. Crypto Briefing, filing at 22:06 Eastern time on 9 November, dated the detection to that day and described the service as still offline with the investigation continuing. The Crypto Times, filing on 10 November, logged the three posts at 04:44, 06:50 and 09:02 India time and reported the service restored. India Standard Time runs five and a half hours ahead of UTC, so the first of those posts falls on 9 November in UTC and the other two on 10 November. Neither outlet published a UTC timestamp for the detection itself.

One line held steady through all three updates. NEAR Protocol and the company's other services were unaffected, NEAR AI said each time. That is the company's own assessment, and nothing independent has been published for or against it.

The claim resting on the hardware

NEAR AI's answer to the breach is the architecture it sells. Customer prompts and inference workloads run inside trusted execution environments, enclaves built into the processor that are meant to keep data away from whoever operates the machine. The company named Intel's Trust Domain Extensions and NVIDIA's confidential computing as the parts doing that work. Responses can carry hardware-backed attestation evidence, which lets a customer check which environment ran a job, and NEAR AI has described an Intel Trust Authority integration as an independent layer over that evidence. Crypto Briefing reported the pitch goes one step further than most: the design is meant to hold data back from the cloud provider and from NEAR AI itself.

Crypto Briefing also placed the incident shortly after a cybersecurity initiative the company launched on 29 October, offering uncensored models for security research under the same standards. No other outlet this desk read draws that connection, so it stands on one account.

What has not been said

The distance between a breach and an exposure is where this story sits. NEAR AI has published no finding that customer data went untouched. It said it replaced affected credentials without identifying them or saying whether the intruder used them. How the access happened and which internal systems were reached are both unpublished. A full incident report is promised. The Crypto Times put the limit plainly: an enclave protects what runs inside it and says nothing about credentials, account records or operational data held elsewhere.

NEAR has had a bad quarter on disclosure. NEAR Intents, a separate product, pinned a $3.8M loss on 23 failed deposit calls in a postmortem of its own. The habit of publishing an incident without a count is not NEAR's alone, and Gnosis Pay disclosed a breach and never named a user figure.

The report is the thing to wait for

A restored service does not close the matter. What NEAR AI publishes next has to say which systems the intruder reached, how the server vulnerability was exploited and whether the enclaves held as described. The Crypto Times suggested customers review their own API-key activity and consider rotating credentials, calling that a general precaution and not a sign that NEAR AI has reported any key compromise. Until the report lands, the strongest statement available is still the company's own.