NEAR AI Discloses SGLang Security Incident and Revokes Exposed API Keys

TL;DR: NEAR AI has published a security incident report involving an SGLang vulnerability in legacy deployments. The company says it revoked exposed credentials and addressed the issue, while acknowledging that some requests could have been observed. Its findings distinguish confirmed exposure from data it has no evidence was accessed.

A security disclosure from NEAR AI highlights a familiar danger in the expanding AI-agent industry: sophisticated applications can inherit serious problems from the infrastructure they run on.

In a November 9 incident report, the organization described an SGLang remote-code-execution vulnerability affecting legacy deployment infrastructure. The investigation identified roughly 100 user details and 18 agents in the affected context, with two Cloud API keys exposed and subsequently revoked.

The company said there was no evidence that stored Private Chat content had been compromised. It also cautioned that some requests made on November 8 could potentially have been observed. Those are not contradictory statements: investigators may find no evidence of one category of access while being unable to rule out another.

The problem was in the service stack

AI applications are assembled from multiple components: model servers, request routers, authentication systems, logging and the user-facing product. SGLang is part of the infrastructure used to serve language models. A flaw at that level can create opportunities for attackers before a user ever interacts with an onchain wallet or signs a transaction.

Remote-code-execution issues can be particularly serious because they may allow an attacker to run unauthorized instructions in the affected environment. The scope of what follows depends on the privileges, network access and data available to that environment.

For readers, this means the incident should not be conflated with a vulnerability in the NEAR blockchain consensus rules or a general compromise of all NEAR wallets. The report concerns particular AI-serving infrastructure and its potential exposures.

What NEAR AI says it did

According to the published account, the team investigated affected deployments, revoked the two exposed API keys, remediated the vulnerability and notified impacted parties. Those are necessary containment steps, though the effectiveness of remediation depends on whether other pathways remain open and whether credentials are rotated wherever they were used.

The report’s numerical counts also need careful interpretation. Approximately 100 user details and 18 agents do not mean 100 users lost funds or that every associated request was read. Exposure, confirmed access and financial loss are separate findings that should not be collapsed into one sensational claim.

Why crypto and AI security overlap

Crypto firms are integrating AI into development, support, trading and operational workflows. An AI tool may eventually handle private business information even when it cannot directly sign onchain transactions. That raises the value of conventional safeguards: isolating services, limiting credentials, monitoring access and maintaining incident-response plans.

The broader institutional push toward blockchain infrastructure, including the NYSE’s tokenized-securities plans, depends on more than smart-contract security. Offchain services must also be robust. A well-audited ledger cannot compensate for a compromised application server.

Likewise, rising institutional attention to crypto assets, which NewsBTC tracks through developments such as Ether ETF flows, makes operational controls increasingly consequential. These links are contextual rather than a claim that those products were involved in this incident.

What users should do

Affected users should rely on NEAR AI’s direct notifications and published guidance. Where they have used credentials in related services, reviewing access permissions and rotating keys is prudent. Users should not assume that ordinary NEAR wallet keys have been exposed simply because the company reported an infrastructure flaw.

The most important follow-up is a clear technical postmortem explaining affected versions, root cause and validation of the fixes. For companies building AI agents around finance, the episode is a reminder that infrastructure security and candid disclosure are both part of earning trust.

Written by the News Desk and edited by Samuel Rae.