NEAR AI pauses cloud inference after detecting unauthorized access

NEAR AI is the AI arm of NEAR Protocol, pitching its Cloud inference and Private Chat products as a confidential alternative to centralized AI providers, with data handling it has marketed as privacy-preserving by design. The service runs on SGLang, an open-source LLM-serving framework widely adopted since 2024 for its speed. A flaw in that third-party stack, not NEAR's own application code, became the entry point attackers exploited, a familiar pattern where shared infrastructure undercuts claims of isolated, confidential computing.

Roughly 100 users and 18 legacy agents now need to treat their NEAR AI account data as exposed, while the two customers whose API keys leaked must revoke and reissue them before resuming integrations. Developers building on NEAR AI Cloud lost access during the pause and must verify their agents still function correctly post-restoration. Anyone running Private Chat sessions gains some reassurance, since NEAR AI found no sign those conversations or uploaded files were touched, but account-level exposure alone forces a credential review.

A product sold on confidentiality has just demonstrated that its security perimeter depends on a widely shared open-source serving layer it does not fully control. That is not unique to NEAR AI, but it sharpens a real tension: AI platforms courting crypto users on privacy promises are often built from the same commodity infrastructure as any centralized cloud competitor. Separating 'no evidence of access' to Private Chat data from 'confirmed no access' matters, and NEAR AI's wording leaves room for both readings.

Watch whether NEAR AI publishes a full post-mortem detailing how the SGLang flaw was patched and whether an independent audit confirms Private Chat data stayed untouched. Also watch the two customers whose API keys were exposed, since any downstream misuse of those keys would show the breach's real-world impact. A continued pause or a second disclosure would signal the fix was incomplete; a clean audit with no follow-up incidents would support NEAR AI's current account of limited exposure.

NEAR ProtocolNEAR it is the asset of the company or protocol in the event

NEAR Protocol rose 1.1% in the 4 hours after the first headline, after falling 1.4% in the 2 hours before.

47 prices, one every 30 minutes · The move was larger before the first headline than after: an observation, not an explanation. The market also moves on things nobody publishes.