What happened in the NEAR AI security incident?
In its incident report, NEAR AI says an attacker exploited a remote code execution vulnerability in SGLang, the open-source inference engine, to run code on one NEAR AI inference server and then move into wider infrastructure. On November 8, 2026, the attacker could have observed requests processed on that server, including confidential-mode requests.
NEAR AI says about 100 users' account details were exposed, along with chat history, settings and logs for 18 agents on a legacy deployment, and NEAR AI Cloud API keys for two customers that appeared in logs and were revoked. It revoked sessions and admin credentials, replaced agent tokens, took Private Chat offline, deployed a fix to all inference servers and reported the vulnerability to the SGLang project. No CVE ID was given in the report.
Who is affected by the NEAR AI breach?
- About 100 NEAR AI users whose emails, names and some plan and balance details were exposed.
- 18 agents on a legacy, non-TEE NEAR AI Agents deployment, whose chat history, settings and logs were exposed.
- Two NEAR AI Cloud customers whose API keys appeared in logs; NEAR AI revoked them.
How to check if you're affected
- Look for a message from security@near.ai, which NEAR AI used to contact affected users.
- If you self-host SGLang, find every deployment and its version, and watch the SGLang project for an advisory on this flaw.
- Check whether your inference servers can reach other infrastructure, credentials or logs that hold API keys.
What to do now
- Follow the account-specific steps in NEAR AI's email.
- Be wary of unsolicited messages that reference your NEAR AI account or balance.
- Upgrade self-hosted SGLang as soon as a fixed release is published, and isolate inference servers from the rest of your environment.
Why inference servers are now a target
Inference engines take untrusted input, prompts and request payloads, from the internet by design, and they often run with access to GPUs, model weights and the credentials that call other services. A code execution bug in that layer is a code execution bug in a public-facing service. Treat open-source inference engines like any other internet-facing dependency: pin versions, follow upstream advisories, run them with the least access they need, and keep API keys out of the logs they write.
Where Nullify fits
How Nullify Covers This Class of Attack
Nullify moves actively exploited vulnerabilities in the dependencies and images you use up the backlog as soon as they're disclosed, and prepares the version change as a pull request for each owner to review.
The vulnerable component was an open-source inference engine running on NEAR AI's servers. If you self-host SGLang, track its advisories like any other dependency.
Frequently Asked Questions
What happened in the NEAR AI SGLang breach?
An attacker exploited a remote code execution flaw in SGLang, the open-source inference engine, on one NEAR AI server, then moved into wider infrastructure. NEAR AI disclosed it on November 9, 2026, fixed the flaw on all inference servers and rotated credentials.
Is SGLang vulnerable?
NEAR AI says it exploited a remote code execution vulnerability in SGLang and reported it to the project. Its report gives no CVE ID, so SGLang users should watch the project's advisories and upgrade when a fixed release ships.
What data was exposed in the NEAR AI breach?
About 100 users' account details, chat history and logs for 18 legacy agents, and two Cloud customers' API keys found in logs. Requests on the affected server, including confidential-mode requests, could have been observed on November 8.
Update Log
- Nov 17, 2026: page published from NEAR AI's incident report. No CVE ID published yet.