NEAR AI hack: data of around 100 users and 18 AI agents exposed

What is confirmed: in the NEAR AI hack, an attacker exploited a flaw in the inference software SGLang to obtain the names and email addresses of around 100 users, in some cases also their plan and account balance, along with chat histories, settings and logs from 18 AI agents. The NEAR AI incident report makes no mention of stolen coins, wallets or private keys; the target was the company's AI service, not the NEAR Protocol blockchain. Our assessment follows further down: the biggest risk for users now comes from phishing messages built on exactly this data.

NEAR AI published the report on November 9, and the trade outlet Decrypt picked it up on November 10. If you have a NEAR AI account, run agents through the service or simply hold NEAR tokens, this piece sets out what is known, what remains open and which steps make sense now.

What happened in the NEAR AI hack via the SGLang flaw

The way in was a remote code execution vulnerability, meaning a flaw that lets an outsider run their own code on a machine from afar. It sat in SGLang, the open-source software NEAR AI uses to run its language models on its own servers. The attacker was able to execute commands on a single inference server and work their way further into the company's infrastructure from there.

According to the report, the intruder reached an internal system behind Private Chat, a feature NEAR AI had already discontinued. The attacker also accessed 18 agents on an older deployment of NEAR AI Agents. These agents were not running in the confidential computing environment the company otherwise promotes.

After discovering the breach, NEAR AI says it revoked all active sessions and administrator access, replaced every agent token, took Private Chat offline and patched the SGLang flaw on all inference servers. All remaining credentials have since been rotated, a forensic review has been completed and access to NEAR AI Cloud has been restored. The company does not say which SGLang vulnerability was exploited. It says it has reported it to the SGLang developers.

Which data the attacker took from NEAR AI

The report distinguishes between confirmed access and a residual risk. These records are confirmed:

• Account data: email address and name of around 100 users, and for some of these accounts also the subscribed plan and the balance held with the service.
• Agent data: chat history, settings and logs of 18 agents on the older agent platform.
• API keys: the logs contained the NEAR AI Cloud API keys of two cloud customers. An API key is a digital access code that programs use to log in to a service. NEAR AI has revoked both keys; the report leaves open whether the attacker used them.

The residual risk concerns November 8. On that day the attacker could have observed requests being processed on the compromised server at the time, including those in confidential mode. NEAR AI says it found no sign of access to stored Private Chat conversations or uploaded files.

One point matters for reading the account balance correctly: it refers to credit for the AI service, in other words what a user has deposited with NEAR AI for model calls or a subscription. A balance in this sense is not a wallet address and not a key. It does, however, tell a fraudster who is a paying customer, and that is precisely what makes the data valuable for targeted deception.

Are NEAR coins and wallets affected by the security incident?

Based on everything NEAR AI and Decrypt report, no. The incident played out on servers that run language models. The NEAR blockchain, its validators and its smart contracts do not appear in the report, and neither do seed phrases, private keys or drained tokens. If you hold NEAR in your own wallet or on an exchange, there is no need to move anything.

The structure helps draw the line. NEAR Protocol is a layer-1 blockchain with its own token, which we describe in detail in our guide NEAR Protocol explained: sharding, staking and the use of the NEAR token. NEAR AI is the AI branch around this ecosystem; the service is operated by Jasnah Inc. and sells private AI computing power. A leak at NEAR AI is therefore a leak at a service provider, comparable to a data incident at a hosting company, and not a fault in the protocol.

The case at the end of September was different. Back then, NEAR Intents halted withdrawals after an exploit of around $3.8 million, as cryptoticker.io reported on October 1 in NEAR Intents halts withdrawals after $3.8 million exploit. That involved funds in an application within the ecosystem. The current incident involves personal and usage data.

The market shows no move of its own. According to CoinGecko, NEAR traded at €3.19 on Tuesday evening, about 2 percent below the previous day and a good 37 percent below its level 30 days ago. That decline started long before the report and tracks the weak overall market; in early November we had already classified it as a price effect in NEAR Protocol falls to $3.70.

What SGLang is and why an inference engine became the entry point

SGLang is an open-source inference engine. Inference is the step in which a fully trained AI model takes a request and produces an answer. The engine loads the model into the memory of the graphics cards, distributes the requests and delivers the answers. Many AI providers use SGLang because the software is fast and supports open models directly.

That very position makes it sensitive. An inference engine sees every request in plain text, because the model has to process the text. Anyone who can run code on that server therefore sits at the point where confidential data is necessarily exposed.

The problem is not new. Decrypt points out that the CERT Coordination Center at Carnegie Mellon University flagged vulnerabilities in SGLang back in July, including flaws that, depending on the server configuration, could allow commands to be executed or credentials to be read. Whether the attacker used one of these flaws is unclear, because NEAR AI does not name the specific vulnerability.

Confidential computing and TEE: where hardware isolation ends

NEAR AI's business model rests on confidential computing. Calculations run in a trusted execution environment, or TEE for short: an area sealed off by the processor that not even the server operator can look into. Prompts are meant to stay encrypted and isolated while the model processes them. According to Decrypt, Brave's browser assistant Leo also uses this infrastructure for one supported model; the report does not, however, list Brave users among those affected.

The incident shows the limit of this approach, and NEAR AI names it itself. Confidential computing protects data from the operator and the host, the company writes, but it does not protect against a vulnerability in the software running inside the environment. That is exactly what happened: the SGLang flaw let the attacker execute code inside an inference server, at the very point where requests are processed in decrypted form.

An analogy makes this tangible. A TEE is like a vault with an armoured door. The door keeps out anyone coming from outside. But if an employee inside lets a stranger control them remotely, the door is useless. In this picture, SGLang is the employee.

NEAR AI says it will in future audit the inference software and the services around the isolated environment as strictly as the environment itself. According to the company, the code, dependencies and configuration of its confidential inference remain public so that third parties can verify what runs there.

Phishing after the data leak: the typical scams with a NEAR AI angle

Phishing refers to fraudulent messages that pose as a trusted sender in order to obtain login details, keys or payments. With a name, email address, plan and account balance, scammers have exactly the material that makes such a message credible. NEAR AI therefore explicitly advises treating unsolicited messages that refer to your account or balance with caution. The company has contacted affected users directly, and the owners of the 18 agents individually.

Data leaks at other crypto services have produced a few well-known patterns. They tend to resurface after incidents like this one:

• The supposed security check: an email styled like NEAR AI asks you to re-verify your account via a link, or your balance will be frozen.
• The compensation offer: affected users are promised restitution in NEAR tokens and asked to connect their wallet or sign a transaction for it. With such a signature, victims often unknowingly grant approval over their tokens.
• The support call: a supposed staff member correctly states your name and plan and asks for a confirmation code or your seed phrase.
• The migration: after the incident, all agents or balances supposedly have to move to a new address supplied in the message.

None of these scams works if you stick to one simple rule: no reputable provider will ever ask for your seed phrase or private key, whether by email, phone or chat.

How to protect your NEAR account and seed phrase after the incident

The following steps take little time and cover the realistic consequences of the data leak:

Contact NEAR AI only through the official website

Open the NEAR AI website yourself via the address bar or a saved bookmark instead of following links in messages. That is where the incident report is published and where you reach genuine support. With every email, check whether the sender address really belongs to the provider's domain.

Renew your password and API keys

If you use NEAR AI Cloud, it is best to create new API keys and delete the old ones, even if your account is not one of the two named. If you have used your NEAR AI password elsewhere, change it there too and enable two-factor authentication wherever the service offers it.

Keep wallet and seed phrase separate

The seed phrase is the sequence of words from which all keys of a wallet can be restored. It belongs offline on paper or a metal plate, never in a chat, an AI assistant or a cloud document. Anyone holding larger NEAR amounts is better off storing them on a hardware wallet; our hardware wallet comparison shows which devices support NEAR and what they cost.

Review signatures and approvals

Do not sign any transaction whose purpose you do not understand. If you have responded to a suspicious message in recent days, check in your wallet's explorer which approvals and access keys sit on your NEAR account, and remove any unknown entries.

GDPR rights after the NEAR AI incident: access and complaints

For users in Germany, the General Data Protection Regulation, or GDPR, generally also applies to providers outside the EU if they offer their services to people in the EU. NEAR AI says it will make the required notifications to authorities; the report does not say which supervisor this concerns.

Regardless of that, Article 15 GDPR gives you a right of access: you can find out from the provider what data it holds about you and whether it was caught up in the incident. Under Article 17 you can request deletion if you no longer use the service. If you receive no answer or an inadequate one, you can lodge a complaint with the data protection authority of your federal state. A written request by email to support is enough; keep the reply.

A claim for damages under Article 82 GDPR requires concrete harm. With around 100 people affected and no sign of financial losses, the scope is narrow; anyone who actually loses money through a phishing message using the stolen data should document the case and report it to the police.

NEAR Protocol after the Intents exploit and the AI data leak: our assessment

For the NEAR ecosystem it is the second security incident within six weeks, even though the two cases have nothing to do with each other technically. The Intents exploit hit funds in an application, the current incident hit data at an AI service.

Our assessment: the damage is limited, the effect on trust in the promise of private AI is not. It is confirmed that the number of people affected is small, at around 100 accounts and 18 agents, that no tokens were drained and that NEAR AI disclosed the cause, scope and countermeasures within a short time. It is equally confirmed that the company sells privacy as its core product and that requests in confidential mode may nevertheless have been visible on November 8. For investors, this means the report contains no reason to sell NEAR tokens because of this incident. Arguing against that reading: the investigation does not quantify the exact number of observed requests, and any link to the price trend would only show up over weeks. This is not a buy recommendation; crypto assets can fall all the way to a total loss.

Other factors remain decisive for the price: the state of the overall market, activity on the blockchain and demand for NEAR applications. Our NEAR price prediction collects scenarios on this. How quickly attacks on devices and wallets are currently piling up is also shown by the iOS flaw from the start of the week, which we described in iOS vulnerability worse_query.

NEAR AI data leak: the danger now arrives by email

The incident itself is contained; the consequences for those affected are only beginning. Three steps are enough to avoid phishing and costly mistakes:

1. Check the trading venue: if you buy or sell NEAR, do it on an exchange with EU authorisation under MiCA rather than via links in messages. Our comparison of regulated crypto exchanges shows which providers hold a licence.
2. Keep keys offline: store larger holdings on a device that never exposes the keys to the internet. Models and prices are listed in the hardware wallet comparison.
3. Secure your everyday wallet: for smaller amounts, an app wallet with PIN, biometrics and a carefully noted seed phrase is enough. The software wallet comparison shows which apps support NEAR.

(As of November 10, 2026. This article is not investment advice. Prices and fee structures change; check the terms with the provider before you buy.)