Quick Answer
NEAR AI says an attacker exploited a remote-code-execution flaw in SGLang on November 8, 2026, exposing account data for approximately 100 users and data from 18 legacy agents. The company says it contained the intrusion, restored NEAR AI Cloud, revoked affected credentials, and patched inference servers. Affected users should treat any NEAR AI-related credentials as sensitive and watch for targeted phishing.
Key Takeaways
- NEAR AI disclosed the intrusion on November 9, 2026, after detecting unauthorized access to part of its infrastructure.
- The attacker exploited a remote-code-execution vulnerability in the open-source SGLang inference engine on one inference server.
- Names, email addresses, and some account plan and balance information were downloaded for about 100 users.
- Data from 18 legacy NEAR AI Agents included chat history, settings, and logs that exposed API keys for two Cloud customers.
- NEAR AI revoked sessions and credentials, replaced agent tokens, patched inference servers, and restored NEAR AI Cloud.
What happened in the NEAR AI security breach?
The NEAR AI security breach began when an unauthorized party exploited a remote-code-execution vulnerability in the open-source SGLang inference engine on one inference server on November 8, 2026. Remote-code execution means an attacker can run commands on a vulnerable system from a distance, which can turn a software flaw into access to the data and processes handled by that server.
NEAR AI disclosed the incident on November 9, 2026, saying it had contained the unauthorized access and restored NEAR AI Cloud. NEAR AI’s incident report identifies SGLang as the entry point and describes the company’s containment actions. The most important point is that the incident involved an inference server, which is the part of an AI service that processes requests and produces model responses.
NEAR AI said the accessed server could have allowed the intruder to observe requests being processed there, including confidential-mode requests. The company also said it found no indication that stored Private Chat conversations or uploaded files were accessed. That distinction limits what NEAR AI says it found, but it does not eliminate the privacy concern for people whose prompts or account data may have passed through the affected infrastructure.
What data did the NEAR AI breach expose?
The NEAR AI breach exposed names and email addresses, along with plan and balance information for some accounts, affecting approximately 100 users. The company said the attacker downloaded that information, which creates a meaningful phishing risk because names, email addresses, and service details can make fraudulent messages appear more credible.
NEAR AI also said the incident reached 18 legacy NEAR AI Agents that were not operating in the company’s confidential-computing trusted execution environment, or TEE. A TEE is a protected hardware-backed area intended to isolate sensitive processing from other parts of a system. The affected legacy agents contained chat history, settings, and logs, according to the company’s disclosure.
The logs exposed API keys belonging to two NEAR AI Cloud customers, and NEAR AI said those keys were revoked. API keys act like passwords for software services, so exposed keys can allow unauthorized use of an account or connected resources until they are disabled and replaced. Organizations that use AI tools should maintain an inventory of active keys, since an incident response is faster when administrators know which credentials connect to each service.
| Data or system affected | What NEAR AI says was involved | Why it matters |
|---|---|---|
| Approximately 100 user accounts | Names, email addresses, and some plan and balance information | Personalized phishing attempts can use accurate account details. |
| 18 legacy NEAR AI Agents | Chat history, settings, and logs | Agent data can contain sensitive context about user activity or workflows. |
| Two Cloud customers | API keys exposed in logs | API keys can permit access until they are revoked and replaced. |
| Private Chat storage | No indication stored chats or uploaded files were accessed | The finding limits the confirmed scope, but does not rule out observed in-flight requests. |
Why does an SGLang flaw matter for AI cloud privacy?
An SGLang vulnerability matters because an inference engine handles the requests that users and applications send to an AI service. A failure in that layer can expose more than account records. It can create a path to view, alter, or interfere with requests while the server processes them, depending on the access achieved and the service architecture.
NEAR AI said the intruder could have observed confidential-mode requests processed on the accessed server. The company’s statement does not say that stored Private Chat conversations or uploaded files were accessed, and it says the company found no indication that they were. For users, the practical distinction is between information stored in a service and information submitted while a request is actively being handled.
A separate Cloudflare disclosure shows why isolation failures around AI-serving infrastructure deserve close attention. Cloudflare said 155 Workers AI responses could contain data from other customers’ requests in an SGLang-related incident. Cloudflare’s technical disclosure describes a separate event, not the NEAR AI breach, but it illustrates how inference-layer problems can cross customer boundaries when request isolation fails.
Users should avoid placing passwords, payment information, private keys, or unnecessary personal records into AI prompts, even when a service offers privacy-oriented settings. Confidential modes can reduce exposure through technical controls, but they do not make a cloud service immune to software vulnerabilities or operational failures.
How did NEAR AI respond to the intrusion?
NEAR AI says it responded by revoking active sessions and administrator credentials, replacing agent tokens, taking Private Chat offline, patching every inference server, and rotating remaining credentials. Those actions address different stages of an intrusion: session revocation removes existing access, credential rotation invalidates secrets, and patching aims to remove the exploited weakness.
NEAR AI also said it revoked the API keys exposed in logs for two Cloud customers. Revoking a key prevents it from continuing to authenticate requests, but a replacement key still needs to be stored securely and configured in the customer’s systems. The practical response for any customer that used NEAR AI Cloud is to review service integrations and confirm that applications are using current credentials.
The company restored NEAR AI Cloud after containing the unauthorized access. Service restoration is important for availability, but it does not by itself establish that all potential downstream effects have been identified. NEAR AI’s finding that it had no indication of access to stored Private Chat conversations or uploaded files is useful scope information, although users should continue to follow any direct notices from the company about their accounts.
AI service incidents also reinforce why organizations should prepare credential rotation procedures before an emergency occurs. The response is more manageable when teams know which applications rely on a token, which account owns it, and how a replacement credential can be deployed without interrupting critical work.
What should affected NEAR AI users do now?
NEAR AI users should review email carefully, change any reused passwords, and confirm that their account recovery details remain accurate. The reported data includes names and email addresses for approximately 100 users, which means fraudulent messages can be tailored to resemble real NEAR AI notifications. Do not use a link in an unexpected email to sign in or reset a password.
- Open NEAR AI through a known bookmark or manually typed address, then review account notices and recent activity.
- Change your NEAR AI password if you reused it on another service, because password reuse can turn one account incident into access to unrelated accounts.
- Enable multi-factor authentication where the service offers it, using an authenticator app or security key rather than relying only on SMS when possible.
- Review active API keys, agent tokens, and connected applications if you use NEAR AI Cloud or legacy agents.
- Replace any credentials that were stored in prompts, logs, agent settings, or connected development environments.
NEAR AI customers who manage production applications should stop and contact their internal security team or NEAR AI support if they cannot identify where an API key or agent token is used. Rotating a credential without updating the dependent application can cause an outage, while leaving a potentially exposed credential active can create a continuing access risk. The appropriate action depends on the account’s configuration and the sensitivity of the connected data.
How can users recognize phishing after the NEAR AI breach?
Phishing messages after the NEAR AI breach are most likely to use urgency, account warnings, or fake credential-reset requests to persuade recipients to disclose passwords or payment information. Accurate names, email addresses, and account-plan details can make a scam look more convincing, but those details do not prove that a message is legitimate. The Cybersecurity and Infrastructure Security Agency’s phishing guidance recommends recognizing and reporting suspicious messages rather than responding to them.
NEAR AI users should verify messages by opening the service directly rather than using an email link or attachment. Check the sender address carefully, and do not provide a password, recovery code, API key, or multi-factor authentication code to a person claiming to offer support. Legitimate support teams do not need a user’s current password to investigate an account.
The breach also provides a useful reminder for teams that use AI services in business workflows. A recent password hashes and live tokens breach showed how exposed authentication data can expand the risks of a security incident. Segregating credentials by application and using limited-scope keys reduces the potential impact when one token must be revoked.
Users who believe an attacker has accessed an account should change the password from a trusted device, revoke unfamiliar sessions, and contact the service’s support channel through its official website. Users who see unauthorized financial activity should also contact the relevant bank or payment provider immediately.
What does the breach show about confidential AI systems?
The NEAR AI incident shows that confidential-computing protections can reduce exposure but do not remove the need to secure the software running around them. NEAR AI said the affected 18 legacy agents were not running in its confidential-computing TEE environment, while the accessed server could have processed confidential-mode requests. The security outcome depends on the full stack, including software patches, credentials, logging practices, and server isolation.
Confidential computing generally aims to protect data while it is being processed by isolating it in protected hardware. That protection can be valuable for sensitive workloads, but a vulnerability in an inference engine can still affect the systems that route requests, manage sessions, or record logs. The most sensible approach is to treat confidential mode as one control in a broader security design rather than as a guarantee of absolute privacy.
Users should also consider what data an AI workflow truly needs. Teams can reduce risk by removing secrets from prompts, using redacted test data, limiting API-key permissions, and keeping logs free of credentials. The same principle applies to consumer tools, where private records should not be entered unless the service and its retention practices are understood.
AI security incidents are not limited to cloud infrastructure. The Android TV box botnet case illustrates how compromised technology can create broader risks when devices or services lack strong controls. The practical lesson is consistent: update software promptly, minimize unnecessary access, and review what information a connected product can collect or transmit.
What is the current status of NEAR AI Cloud?
NEAR AI Cloud is restored after the company said it contained the unauthorized access to part of its infrastructure. NEAR AI disclosed the restoration on November 9, 2026, alongside its statement that it had patched every inference server and rotated remaining credentials. NEAR AI’s security update remains the primary source for changes to the company’s assessment.
NEAR AI took Private Chat offline as part of its response, although the supplied disclosure does not establish the current availability of that specific feature after the broader cloud restoration. Users should check the company’s account notices and service status information before relying on a feature for sensitive work. Availability can change as an investigation continues or as technical safeguards are updated.
For most users, the immediate priority is account hygiene rather than service migration. Review credentials, remain alert for phishing, and avoid submitting unnecessary secrets through AI tools. Organizations that use multiple AI services should also document which teams use each platform, since service inventories make it easier to evaluate exposure when a provider discloses an incident.
FAQ
Did the NEAR AI breach affect all users?
NEAR AI says the downloaded account data affected approximately 100 users, not all users. The company also said the intrusion reached 18 legacy NEAR AI Agents, although the supplied disclosure does not state how many people used those agents.
Did attackers access stored NEAR AI Private Chat conversations?
NEAR AI says it found no indication that stored Private Chat conversations or uploaded files were accessed. NEAR AI also said the intruder could have observed requests being processed on the accessed server, including confidential-mode requests.
Were API keys exposed in the NEAR AI breach?
NEAR AI says logs from the affected legacy agents exposed API keys for two Cloud customers. The company said it revoked those keys, but customers should still verify that connected applications now use replacement credentials.
What should NEAR AI users do after the breach?
NEAR AI users should change any reused password, review account activity, and treat unexpected account emails as possible phishing. Cloud customers should also review API keys, agent tokens, and connected applications for credentials that need replacement.
Is the NEAR AI breach related to the Cloudflare Workers AI incident?
The NEAR AI breach and Cloudflare incident are separate disclosures involving SGLang-related AI infrastructure risks. Cloudflare said 155 Workers AI responses could contain other customers’ request data, while NEAR AI described unauthorized access to one inference server.