NEAR AI restores Cloud service after SGLang security flaw

What happened

NEAR AI has restored its Cloud inference service after suspending it when an attacker gained unauthorized access to its systems, according to reports from Crypto.news and NewsBTC. The company published a security incident report on November 9, one day after the attack on November 8. It said engineers applied fixes across its inference servers and replaced affected credentials before bringing Cloud operations back online.

The entry point was SGLang, an open-source inference engine used on NEAR AI's servers. The attacker exploited a remote code execution vulnerability to run code on one inference server, the company said. From there, the attacker reached an internal system connected to the previously discontinued Private Chat service and downloaded account information for approximately 100 users. The records included email addresses and names, and for some users subscription plan and account balance details. NEAR AI said it found no evidence that stored Private Chat conversations or uploaded files were accessed.

The review also confirmed access to 18 agents running an older version of NEAR AI Agents. Those deployments were not operating inside the company's trusted execution environment, or TEE, which provides hardware-based isolation. The data retrieved included chat histories, configuration settings and operational logs. Two NEAR AI Cloud customers had API keys exposed through those logs, and both credentials were revoked. The company further said the attacker could have observed requests processed on the compromised inference server on November 8, including requests in confidential mode, but it did not say how many.

In response, NEAR AI revoked active sessions and administrative credentials, replaced agent tokens and took Private Chat offline. It reported the SGLang vulnerability to the project's maintainers and said its security reviews would now cover inference software, related services, dependencies and deployment configurations. The company completed a forensic review and contacted affected users by email from [email protected], while owners of the 18 compromised agents received individual notices. NEAR AI said regulatory notifications would follow where required.

Why it matters

The incident highlights a risk that runs through AI-agent services built on crypto infrastructure: an application can inherit serious weaknesses from the model-serving software underneath it. SGLang sits in the inference layer, and a flaw there can create exposure before a user ever interacts with an onchain wallet or signs a transaction, even when blockchain consensus rules are untouched.

NEAR AI's account also draws a line between different kinds of harm. Roughly 100 user records and 18 agents are exposure and confirmed access figures, not evidence that 100 users lost funds. NewsBTC noted that exposure, confirmed access and financial loss are separate findings. For crypto firms adding AI to support, development or operations, the episode underlines conventional controls such as service isolation, limited credentials, access monitoring and incident-response planning.

What the data shows

The confirmed figures in NEAR AI's disclosure are specific but limited: about 100 user account records, 18 legacy AI agents and two Cloud API keys. The user records included names and email addresses, with plan and balance data for some accounts. The agent data included chat histories, settings and logs. NEAR AI did not specify how many requests may have been observed on the compromised inference server. It also said it had no evidence that stored Private Chat conversations or uploaded files were accessed.

Background

SGLang is an open-source inference engine used to serve language models, and NEAR AI's confidential computing products had been expanding before the incident. In October, the platform surpassed one million NEAR staked through a service that lets users fund AI inference with staking rewards, according to Crypto.news. NEAR co-founder Illia Polosukhin warned that month about software vulnerabilities across internet infrastructure and called for stronger AI-assisted security systems.

The incident occurred in the same week that investigators identified NEAR-based tokens being used to move funds tied to the Binance exchange hack. Researchers found roughly 117,000 NEAR routed through pools involving tokens created between November 5 and November 7. Neither NEAR AI's report nor the researchers' findings established a connection between the two events, and NEAR AI said the NEAR Protocol blockchain and other NEAR services were unaffected.

What is still unclear

One open detail is the scale of potential request observation. NEAR AI said requests on the compromised inference server could have been seen, including confidential-mode traffic, but it did not provide a count. NewsBTC said the most important follow-up would be a clear technical postmortem explaining affected versions, the root cause and validation of the fixes.

What to watch

Watch for NEAR AI's follow-through on broader security reviews covering inference software, dependencies and deployment configurations, as well as any update from SGLang's maintainers. Also watch whether affected users raise further issues and whether regulatory notifications are made where required. NewsBTC noted that the effectiveness of remediation depends on whether other pathways remain open and whether credentials are rotated wherever they were used.

Questions readers ask