Skip to content
CRYPTO & DECENTRALIZED TECH

NEAR AI Details SGLang Vulnerability in Legacy Infrastructure Affecting Select Deployments

NEAR AI has published a comprehensive security incident report detailing a vulnerability found within legacy deployments of SGLang, a critical component of modern language model serving infrastructure. The disclosure highlights the unique operational and security challenges that emerge as sophisticated artificial intelligence applications interface with complex underlying software stacks. According to the company, the issue has been successfully addressed, exposed credentials were promptly revoked, and affected parties were notified, though investigators note that certain data requests handled during the active window could potentially have been observed.

The security disclosure underscores a persistent and growing vulnerability vector within the rapidly expanding AI-agent ecosystem. As organizations deploy increasingly autonomous and sophisticated AI applications, these tools inherently inherit the security posture, vulnerabilities, and structural weaknesses of the underlying infrastructure upon which they operate. A flaw at the foundational server or routing level can introduce severe risks before an end user ever engages with higher-level application features, such as cryptographic wallets or blockchain transactions.

In the incident report released on November 9, NEAR AI outlined the nature of the problem, which centered on a remote-code-execution vulnerability tied to legacy deployment infrastructure running SGLang. Following the detection of the issue, the organization initiated an internal investigation to determine the scope and scale of the exposure. The review identified approximately 100 individual user details and 18 distinct agents operating within the affected context. Additionally, investigators confirmed that two Cloud API keys were exposed during the incident, prompting the immediate revocation of those credentials to prevent unauthorized access or further exploitation.

Despite the exposure of specific deployment details and API keys, NEAR AI reported that it found no evidence to suggest that stored Private Chat content had been compromised. At the same time, the company cautioned that some requests processed on November 8 could potentially have been observed by external parties. Security analysts point out that these statements are not contradictory; digital forensics investigations frequently uncover clear indicators of certain categories of system access while remaining unable to conclusively rule out other forms of passive observation or transient data handling.

The problem was located squarely within the broader service stack that powers modern artificial intelligence applications. AI systems are rarely monolithic products; instead, they are assembled from a diverse array of interconnected components, including foundational model servers, request routers, authentication systems, internal logging utilities, and user-facing interfaces. SGLang functions as part of the specialized infrastructure utilized to serve large language models efficiently. When a critical flaw emerges at this level, it can create operational openings for malicious actors well before any user-facing interaction takes place.

Remote-code-execution issues are regarded by cybersecurity professionals as particularly severe because they can permit an unauthorized entity to execute arbitrary instructions within the compromised operating environment. The ultimate impact and scope of such an event depend heavily on the specific privileges, network access boundaries, and local data available to the affected environment at the time of the incident.

For observers and participants in the broader digital asset ecosystem, the incident report makes it clear that the event should not be conflated with any vulnerability in the underlying NEAR blockchain consensus rules or a systemic compromise of NEAR user wallets. The scope of the report is strictly confined to specific AI-serving infrastructure components and their potential operational exposures, rather than the core distributed ledger technology.

According to the published account, NEAR AI undertook several containment and remediation measures upon discovering the issue. The team investigated the affected deployments, successfully revoked the two exposed API keys, remediated the underlying software vulnerability, and directly notified impacted individuals and entities. Cybersecurity experts emphasize that while these actions represent necessary and standard containment steps, the overall effectiveness of any remediation effort depends on ensuring that no secondary or backdoor pathways remain open, and that associated credentials are thoroughly rotated across every service where they might have been utilized.

The numerical metrics provided in the incident report require careful contextual interpretation. The identification of approximately 100 user details and 18 agents does not indicate that 100 users suffered financial losses or that every associated request was intercepted or read by an unauthorized actor. In security disclosures, confirmed exposure, actual unauthorized access, and tangible financial loss are distinct analytical findings that must be evaluated independently rather than collapsed into a single sensationalized narrative.

The incident highlights a broader industry trend regarding the intersection of cryptocurrency operations and artificial intelligence security. As blockchain firms increasingly integrate AI tools into their development pipelines, customer support systems, trading desks, and operational workflows, the stakes surrounding infrastructure security continue to rise. An advanced AI utility may eventually process sensitive corporate or individual business information even if it lacks the direct capability to sign onchain transactions. This dynamic elevates the importance of traditional, robust security safeguards, such as strict service isolation, credential minimization, continuous access monitoring, and comprehensive incident response planning.

The expanding institutional integration of blockchain infrastructure—exemplified by broader market developments such as traditional financial exchanges planning tokenized-securities venues—relies heavily on the security of offchain services as much as on-chain smart contracts. A thoroughly audited and secure public ledger cannot automatically compensate for a compromised application server or weak infrastructure management.

Similarly, rising institutional participation in digital assets, reflected in fluctuating ETF fund flows and shifting market liquidity, places a premium on resilient operational controls. While these macroeconomic and infrastructural developments are distinct from the NEAR AI incident, they illustrate the broader environment in which operational security lapses can carry significant reputational and operational consequences.

For users who may have been affected by the incident, the primary guidance is to rely on direct communications and published notices from NEAR AI. Individuals who utilized credentials across related services are advised to review current access permissions and proactively rotate keys as a standard precautionary measure. Furthermore, experts stress that users should not assume standard NEAR blockchain wallet keys have been compromised simply because an infrastructure-level software flaw was reported within an AI-serving environment.

The most valuable follow-up to such an incident is a transparent and detailed technical postmortem that clearly outlines the affected software versions, identifies the precise root cause, and provides validation that the implemented fixes are effective. For companies and developers building sophisticated AI agents around financial and operational workflows, the episode serves as a timely reminder that rigorous infrastructure security and candid, precise public disclosures are essential components of building and maintaining long-term user trust.