Security
NEAR AI says an attacker exploited an SGLang remote code execution flaw on an inference server, exposing data from about 100 users and 18 legacy agents and possibly confidential requests inside its TEE
NEAR AI says an attacker used an SGLang RCE flaw, taking data on about 100 users and 18 legacy agents. Confidential requests may also have been seen.
NEAR AI disclosed on November 9 that an attacker exploited a remote code execution flaw in SGLang, the open-source inference engine, and ran code on one of its inference servers on November 8. The 100 users and the 18 agents are two separate sets of affected data. The company says it patched its servers and restored its Cloud service.
The essentials
- The entry point: The Hacker News says the attacker ran code on a NEAR AI inference server through the SGLang flaw. SGLang runs inside the company’s trusted execution environment (TEE), which is how requests there could be reached.
- Users: about 100 users’ account details were downloaded, including names and email addresses and, for some, plan and balance information, per Cloud Security Alliance and crypto.news. The attacker moved to systems behind NEAR AI’s deprecated Private Chat feature.
- Agents: data from 18 agents on an older NEAR AI Agents deployment that ran outside the TEE was taken, including chat history, settings and logs. Cloud Security Alliance says API keys for two Cloud customers had been written to those logs.
- Response: Bitcoin.com News says NEAR AI restored Cloud services after patching, and Blockonomi says credentials were rotated. The company acknowledged limits in confidential-computing protection.
What the enclave did and did not protect
Cloud Security Alliance notes that attestation proves which software build is running, not that the build is safe. A flaw inside the enclave can still expose the requests it handles.
Possible, not confirmed
Blockonomi says the attacker may have seen confidential AI requests on one server. The pages read do not show that any specific request was read.
What is not established
- Whether any confidential request was actually viewed: only “may have”, not confirmed.
- Whether the number of affected users will rise: “about 100” is the current count.
- The SGLang vulnerability identifier and fix version: not given in the pages read.
- NEAR AI’s own disclosure was read through relays, not directly.
Summary table
| Point | What is stated as of November 7, 2026 | Source |
|---|---|---|
| Date | Exploited on November 8, disclosed on November 9 | Cloud Security Alliance |
| Users | About 100 users’ names and emails, some plan and balance data | crypto.news |
| Agents | Chat history, settings and logs from 18 legacy agents | The Hacker News |
| Leaked keys | API keys for two Cloud customers were in the logs | Cloud Security Alliance |
| Recovery | Servers patched, credentials rotated, Cloud restored | Bitcoin.com News |
FAQ
Were 118 users or agents hacked?
No. About 100 users’ account details and data from 18 legacy agents are two separate sets.
Did the confidential-computing setup fail?
The flaw was in software running inside the enclave, so the enclave did not stop the attacker from reaching requests there.
Is the flaw fixed?
NEAR AI says it patched its servers. The pages read give no SGLang version details.
Method and limits
Checked on November 12, 2026 against The Hacker News, Cloud Security Alliance, Bitcoin.com News, crypto.news and Blockonomi; NEAR AI’s own statement was read through relays.
Sources
- The Hacker News: SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users (verified 2026-11-08)
- Cloud Security Alliance: Inside the Enclave, Still Exposed: The NEAR AI SGLang Breach (verified 2026-11-08)
- Bitcoin.com News: Near AI Confirms Security Breach Affecting 100 Users and 18 Agents (verified 2026-11-08)
- crypto.news: NEAR AI fixes SGLang security flaw, restores Cloud service (verified 2026-11-08)
- Blockonomi: Near AI Says Hacker May Have Seen Confidential AI Requests in Server Breach (verified 2026-11-08)
Spotted an error in this article? Report it · Our corrections policy