NEAR AI Restores Cloud After Data Leak Affecting Around 100 Users

The company NEAR AI has restored the operation of its Cloud platform following a cyberattack in which the attacker gained access to data of approximately 100 users and 18 AI agents. The incident occurred on November 8, 2026, due to a vulnerability in the SGLang software used for processing requests to artificial intelligence models.

According to a security report dated November 9, the attacker exploited a remote code execution vulnerability on one of the NEAR AI servers. Subsequently, they accessed other components of the internal infrastructure. The company temporarily suspended the operation of NEAR AI Cloud, fixed the vulnerability, and replaced the affected credentials. As of the latest report, the cloud service is once again available to users.

What Data Was Compromised

During the investigation, NEAR AI determined that the attacker accessed an internal system related to the previously decommissioned Private Chat service.

From it, information about approximately 100 users was downloaded, including names and email addresses. For some accounts, data on tariff plans and balances were also accessible. However, the company found no evidence of access to saved Private Chat conversations or files uploaded to it.

Unauthorized access was separately confirmed to 18 AI agents operating on an outdated version of NEAR AI Agents. These agents did not use the TEE secure execution environment designed for hardware isolation of confidential data.

The attacker obtained the agents' conversation history, their settings, and technical logs. The latter contained API keys of two NEAR AI Cloud clients. The company revoked both keys, preventing their further use.

Vulnerability Affected Confidential AI Request Processing

The investigation also revealed a potential risk for requests processed on the compromised server on November 8. According to NEAR AI, the attacker could have accessed their content, including requests in confidential mode. To protect such data, the platform uses trusted execution environments (TEE). This technology isolates computations from the main server system and the infrastructure operator.

However, hardware isolation does not eliminate errors within the software processing the data. The SGLang vulnerability allowed the attacker to execute code directly in this environment.

In the official report, NEAR AI did not specify how many confidential requests might have been viewed. The company also did not disclose information about the attacker's identity.

NEAR AI Fixed the Vulnerability and Restored Cloud

After discovering the attack, NEAR AI specialists terminated unauthorized access to the infrastructure, revoked active sessions and administrative credentials, and replaced AI agent tokens. The SGLang fix was installed on all request processing servers. The company also completed the replacement of other potentially affected credentials and conducted a technical investigation.

Users whose information was compromised received individual notifications. Separate instructions were sent to the owners of the 18 affected agents. NEAR AI recommends paying attention to suspicious messages mentioning account data or user balance, as such information could be used for fraud attempts.

The company also reported sharing information about the vulnerability with SGLang developers and intends to strengthen checks of software, its dependencies, and deployment settings. According to NEAR AI, the NEAR Protocol blockchain and other ecosystem services were not affected. There is no evidence linking this incident to the recent Binance hack.

The cloud platform has been restored, but the exact scope of potentially viewed confidential requests remains unknown.