Skip to content

AI NewsPublished 5 min read

AI Security Incident Exposed NEAR AI Agent Data

nodes passing glowing task tokens along branching paths
Listen to this article · 7:50 · AI-generated narration
0:00 / 7:50
Chapters

The short version

NEAR AI said an attacker exploited a flaw in the SGLang inference engine to download account details for about 100 users and data from 18 legacy agents. The company also said confidential computing does not protect against a vulnerability in software running inside the environment, and that the SGLang flaw let the attacker access requests in processing on one server.

  • Two NEAR AI Cloud customers had API keys exposed in logs, and both keys were revoked, the company said.
  • NEAR AI reported finding no sign that stored Private Chat conversations or uploaded files were accessed.
  • Theblock reported that NEAR AI Cloud access, paused during the investigation, has been restored.
  • Thehackernews reported a Chainstack breach on November 3, 2026, and Shattered reported a Chainalysis breach on November 7, 2026; neither outlet tied those incidents to NEAR AI.

NEAR AI discloses an SGLang breach

NEAR AI said in a security incident report published November 9, 2026, that an unauthorized party exploited a remote code execution vulnerability in SGLang, the open-source inference engine running on its servers. The AI security incident exposed names, email addresses and, for some accounts, plan and balance information for approximately 100 users, NEAR AI said.

According to En Bloomingbit, which published its account on November 10, 2026, the attack also reached 18 older AI agents that were not operating in NEAR AI's confidential computing environment. Chat histories, settings and logs from the legacy agents were exposed, En Bloomingbit reported. Two Cloud customers also had NEAR AI Cloud API keys exposed in logs, and both keys have been revoked, NEAR AI said.

The attack moved inward from one server

The intrusion began on November 8, when the attacker gained code execution on one inference server before moving into other parts of NEAR AI's infrastructure, Crypto reported on November 10, 2026. From there, the attacker reached an internal system connected to the discontinued Private Chat service, Crypto reported.

NEAR AI said that on November 8 the attacker could also have observed user requests being processed on the compromised server, including requests running in confidential mode. In its report, NEAR AI wrote: "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 found no indication that stored Private Chat conversations or uploaded files were accessed, according to its incident report.

NEAR AI revoked credentials and restored Cloud

On detection, NEAR AI revoked all active sessions and administrative credentials, replaced all agent tokens and took Private Chat offline, the company said. It also deployed a fix for the SGLang flaw across all inference servers and reported the vulnerability to the SGLang project, NEAR AI said. Rotation of all remaining credentials and a full forensic review are complete, according to the company.

Theblock reported on November 10, 2026, that access to NEAR AI Cloud, paused during the investigation, has been restored. NEAR Protocol and other NEAR services were not affected, Crypto said.

Affected people have been notified individually, and the company plans to make any required regulatory filings, En Bloomingbit reported.

Chainstack and Chainalysis disclosed breaches earlier

Blockchain infrastructure provider Chainstack disclosed an attack in which an unauthorized party accessed a database of customer account information, including password hashes, Thehackernews reported on November 3, 2026. Chainstack detected the intrusion on October 31, 2026, and cut off the attacker's access on the same day, Thehackernews reported. According to Thehackernews, the Chainstack attack is assessed to share overlaps with a campaign by a China-linked cluster targeting high-value firms with ARTEX, an autonomous AI agent framework. XL.net's AI Desk covered that framework in an earlier report on Censys and ARTEX.

Blockchain analytics company Chainalysis said an external actor used "offensive agentic techniques" to gain unauthorized access to its systems and internal tools, Shattered reported on November 7, 2026. Names, usernames, password hashes and live tokens for everyone on the Chainalysis platform were exposed, Shattered reported.

In its weekly breach roundup dated November 6, 2026, Securityboulevard listed developer platforms that hold AI access tokens among the trusted intermediaries it called the story of that week.

Tron's take

My take: NEAR AI's disclosure is a vendor-diligence story, not a reason to rush into or out of AI tools. That is my reading of the news, not a reported result. Three details from the company's own report stand out to me. The entry point was open-source inference software, which NEAR AI said its confidential computing environment does not protect against. The legacy agents breached in the attack ran on an older deployment outside that environment. And customer API keys ended up in logs.

My advice is to put three questions to every hosted AI vendor a business pays for: which inference engine runs the service and how fast the vendor patches it, whether prompts or API keys can land in logs and who can read them, and whether retired products and old deployments are actually shut down. I would also rotate any API key issued to an AI vendor on a set schedule and right after any vendor breach notice, then confirm the old key no longer works.

I would not add a hosted AI service until its vendor answers those questions in writing. I also would not tune out AI security news; an owner who follows it can time adoption on purpose instead of reacting to hype. XL.net sells security assessments and managed IT, and a vendor and API key review like the one above falls within that work.

Questions I'd expect

What data did NEAR AI say was exposed?

NEAR AI said the attacker downloaded names, email addresses and, for some accounts, plan and balance information for approximately 100 users, plus chat history, settings and logs from 18 agents on a legacy deployment.

How did the attacker get in?

Through a remote code execution vulnerability in SGLang, the open-source inference engine on NEAR AI's servers, according to the company. The flaw gave the attacker code execution on one inference server and access to internal infrastructure from there.

Did confidential computing protect NEAR AI customers?

NEAR AI said confidential computing protects data from the operator and the host but not from a vulnerability in the software running inside the environment. The company said the SGLang flaw let the attacker access requests in processing on one server, including confidential ones, while it found no indication that stored Private Chat conversations or uploaded files were accessed.

Is NEAR AI Cloud available again?

NEAR AI said access to NEAR AI Cloud has been restored after it patched SGLang on all inference servers and rotated credentials.

All AI news