NEAR AI security breach exploited a SGLang vulnerability, revealing user information and legacy agent data before service restoration.
NEAR AI has restored its Cloud service after containing a security breach that exposed account details for approximately 100 users and data from 18 legacy AI agents. The NEAR AI security breach began on November 8, 2026, when an attacker exploited a flaw in the open-source SGLang inference engine, according to the company’s incident report.
Summary
Key takeaways
- An SGLang flaw enabled code execution on one inference server.
- The attacker accessed 18 legacy agents outside the confidential computing environment.
- Two Cloud customers’ exposed API keys have been revoked.
- NEAR AI found no evidence of access to stored Private Chat conversations or uploaded files.
How the NEAR AI security breach exposed account data
A remote code execution vulnerability in SGLang allowed the attacker to run code on one inference server and reach an internal system connected to the previously discontinued Private Chat feature. NEAR AI described that path in its security report published on November 9, 2026.
The attacker downloaded users’ names and email addresses, along with subscription plan and account balance details in some cases.
The intrusion also reached 18 agents on an older NEAR AI Agents deployment. Those agents operated outside the company’s confidential computing environment, and the accessed material included conversation histories, settings and logs.
Logs exposed API keys belonging to two Cloud customers. NEAR AI has revoked both credentials. The company found no indication that the attacker accessed stored Private Chat conversations or uploaded files, and said NEAR Protocol and other NEAR services were unaffected.
Confidential computing did not block the software exploit
The SGLang vulnerability allowed access to requests being processed on the compromised server, including requests in confidential mode. NEAR AI said the attacker could have observed those requests on November 8, 2026.
The company distinguished that exposure during processing from access to stored conversations. It also explained that confidential computing protects against access by operators and hosts, not against flaws in software executing inside the protected environment.
“Confidential computing protects data from the operator and the host. It does not protect against a vulnerability in the software running inside the environment, and that is what happened here,” NEAR AI said.
Following the NEAR AI security breach, the company said its security analysis now applies the same rigor to inference software and services surrounding the enclave as to the enclave itself. It added that its confidential inference stack’s code, dependencies and deployment configurations remain public for verification.
Cloud restored after patches and credential changes
NEAR AI said it blocked the attacker’s access, patched all inference servers and replaced affected credentials before restoring Cloud service. Cloud access had remained paused during the investigation and security work.
The response also included invalidating active sessions and administrative credentials, rotating agent tokens and taking Private Chat offline. The company completed a full forensic review and the remaining credential rotations.
NEAR AI contacted affected users and the owners of the 18 agents individually. Affected customers received instructions from [email protected]. In its November 10, 2026, public update, the company said: “If you haven’t heard from us, no action is needed.”
Read also: Near Protocol wants to build the largest open-source AI model with 1.4T parameters · NEAR Protocol Q3 activity includes $10.4B in cross-chain volume
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
