Disclaimer: Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details.

Near AI Breach Exposes User Details and Data From 18 AI Agents

Near AI Breach Exposes User Details and Data From 18 AI Agents

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

In May 2026, Decrypt reported that a flaw in SGLang, the serving framework that sits underneath a large family of AI agents, let an attacker reach Near AI's internal systems and potentially view confidential AI requests. User details and data tied to 18 AI agents were exposed. The story's own framing is the part that should stop a trader cold: the incident "exposed the limits of hardware-backed privacy protections."

That reads like an infrastructure headline. It is also a headline about the AI trading bot and algorithmic trading platform category, because the same serving frameworks, the same cloud agent orchestration, and the same API credential model now sit underneath a lot of retail trading software. When we run our six-month funded-account trials in 2026, the security checklist we apply to a bot is no longer a footnote. It is a line item that can wipe an account just as fast as a bad drawdown.

Across our 2026 review cycle we have benchmarked a cohort of automated trading products, including Zephyr AI's adaptive engine, against a fixed checklist covering API key custody, withdrawal flow, and how much the vendor actually discloses about data handling. This article walks through what the Near AI breach means for that checklist, and where most AI trading bots are quietly exposed.

What the Near AI breach actually exposed

The mechanics matter more than the headline. Per Decrypt, the entry point was not a stolen password or a phishing email. It was a flaw in SGLang, an open-source framework used to serve and coordinate AI agents at scale. An attacker who reached that layer could get to internal systems and potentially view confidential AI requests, which is to say the raw prompts and payloads that agents were processing.

The exposure touched data from 18 AI agents. In plain terms, a single weak link in the serving stack turned into visibility across a whole fleet of agents. The report's conclusion, that hardware-backed privacy protections have limits, is the sentence every trader running an AI trading bot should underline.

Here is the translation. If you run an algorithmic trading platform that executes through a hosted AI agent, your broker API keys, your position data, and your strategy parameters are flowing through infrastructure you do not control. The Near AI case is a worked example of what happens when that infrastructure fails.

What does an AI trading bot actually expose?

When we map a retail AI trading bot onto the same attack surface that Near AI's agents occupied, three asset classes sit in the blast radius.

The first is credentials. Most cloud-hosted bots store broker API keys server-side so the bot can place orders without you being logged in. That is convenient and it is also the whole problem: a leaked key can place trades, not just read them.

The second is the strategy specification. For a copy trading or social trading platform, the leader's strategy logic and the follower allocation book are both sensitive. For an AI signal provider, the signal generation logic is the product.

The third is account and personal data. Names, emails, KYC documents, and balances. The Near AI report describes user details exposed alongside agent data, which is the same combination a bot vendor holds.

Not every architecture carries the same exposure. Self-hosted frameworks such as Backtrader and NautilusTrader keep strategy logic and execution on the trader's own machine, which shrinks the vendor-side surface to almost nothing. The trade-off is that you own the operational burden. Cloud AI trading bots and robo-advisors invert that: less setup, more of your data sitting on someone else's server.

Why are AI trading bots such a soft target?

The economics are the reason. An AI trading bot that connects to three or four broker APIs is a single point of failure that aggregates credentials across all of them. A copy trading platform adds a second layer, because it holds follower funds-allocation data on top of the leader's strategy.

An attacker who compromises that middle layer need not break into a broker; they inherit the session. That risk differs from MetaTrader's self-hosted expert-advisor model, where an EA runs on a user-controlled VPS and the credential can stay local, though that setup still depends on VPS hardening, broker permissions, and the EA's security posture.

We treat this asymmetry the way we treat leverage: it is fine until the day it is not. And unlike a margin call, a credential breach does not send you a warning first.

How accurate are vendor security claims, really?

This is the security version of the backtest-versus-live gap, and we apply the same skepticism.

Vendors describe encryption at rest, encryption in transit, and hardware-backed protections. The Near AI incident is a reminder that those descriptors describe a design intent, not a guarantee. A flaw in one component of the serving stack was enough to reach internal systems and potentially view confidential AI requests, hardware backing notwithstanding.

Our stance in the 2026 review cycle is simple. We treat a vendor's security page the way we treat a backtest equity curve: as a claim to be verified, not a result to be trusted. Where a provider publishes no data-handling documentation, we log that as a gap rather than a passing grade. Where the documentation exists but the specifics are vague, we note "verify with provider" and move on.

Table 1: what the Near AI incident exposed, layer by layer

Layer What was exposed Retail trading equivalent What to verify with your bot provider
AI agent serving framework Flaw in SGLang allowed access to internal systems The hosted engine that runs your bot's decisions Which serving framework the vendor uses and how it is patched
AI agent data User details and data from 18 AI agents Your account data, positions, and prompts Whether agent data is isolated per user
Confidential AI requests Potentially viewable by the attacker Your strategy parameters and order intent Whether request payloads are logged and for how long
Hardware-backed privacy Limits exposed by the incident Any secure-enclave marketing claim Verify with provider; no register lists this
Credential custody Not specified in the source Broker API keys held server-side Whether keys are scoped, rotatable, and revocable

Source: Decrypt. Fields marked verify with provider are not disclosed in the source material and were not available in our test window.

What a breach does to a real trading account

Most risk conversations about AI trading bots center on drawdown. The Near AI breach points at a second failure mode that no equity curve captures: unauthorized order flow.

If an attacker can view confidential AI requests, the step to placing a trade through a compromised API key is short. For a retail account, the practical damage is not a slow bleed. It is a burst of positions you did not authorize, opened at market, possibly with leverage, at a moment when you are not watching.

That is a drawdown with no stop, no cooldown, and no strategy logic behind it. In our 2026 review cycle we added a credential-revocation test to the checklist for exactly this reason: how fast can a trader cut the bot off if the provider is compromised?

The fee model versus the security you actually get

There is an uncomfortable relationship between what a bot charges and how much it spends on the boring parts of security.

Subscription models spread the cost of infrastructure across users. Performance-fee models align the vendor with your P&L but can underfund fixed costs like security engineering. Free-with-spread-markup models push the cost into execution, which means the vendor's incentive is volume, not disclosure.

None of this is a rule, and we are not going to pretend a higher fee buys safety. But when we compare an AI trading bot's pricing page against its security documentation, the two are frequently written to very different standards. A vendor that publishes a precise fee schedule and a vague data-handling page is telling you something about where its attention goes.

Broker and API integration risk

Every integration is a credential, and every credential is a door.

A bot that reaches your broker through a bridge, a webhook, or a direct API is holding a key that can move money. MetaApi and similar connection layers, MetaTrader bridges, and TradingView webhook setups all sit in this category. The convenience is real. So is the exposure.

The Near AI incident is a reminder that the risk is not only at the broker. It is at every hop between the bot's decision and the broker's order book. A hosting layer, an agent framework, or a third-party service the bot calls for sentiment or news is part of the chain.

What we look for is scoping. A well-built integration uses keys that can trade but cannot withdraw, that can be rotated without rebuilding the account, and that can be revoked in one action. Those are the questions to put to any vendor, and they are the questions the Near AI case makes urgent rather than theoretical.

When the bot does something not in its spec

Strategy deviation is the trading version of the Near AI problem, and it is under-discussed.

An AI agent that calls out to a third-party endpoint to fetch news or sentiment is doing something that may not be in the strategy specification you signed up for. Near AI's agents made outbound confidential requests; that is a data-flow behavior, and data-flow behavior is exactly what a strategy spec should pin down.

For a trading bot, the deviation flag we care about is an outbound call that carries position data or account identifiers to a service the vendor has not disclosed. That is not a strategy tweak. It is a data-handling event, and it belongs in the same risk register as a drawdown breach.

There is a second-order risk the source material did not spell out, and it is the one that should worry portfolio managers most. If 18 agents were reachable through one serving framework, those agents may share more than infrastructure. They may share strategy DNA. When a cohort of bots is built on the same model stack and fed the same signal inputs, a breach that exposes their request payloads can expose correlated positioning across many accounts at once. That is not just a privacy problem. It is a crowding problem, because the same event that leaks the data can reveal how much of the market is leaning the same way. A single-user data leak is a personal loss. A framework-level leak across a bot cohort is a market-structure event, and almost no retail bot vendor prices that risk into its documentation.

Can you actually stop an AI trading bot cleanly?

This is the question almost nobody asks until it matters, and it is where the Near AI story gets practical.

If your bot provider is compromised, your exit path is: revoke the broker API key, close open positions, and withdraw the balance. Each of those steps depends on infrastructure that may be the same infrastructure that just failed.

In our review framework we score disengagement as a first-class feature. A provider that lets you self-serve key rotation and withdrawal without a support ticket is materially safer than one that gates both behind a human. This is one dimension where Zephyr AI's withdrawal and account-control flow sets the benchmark we hold the rest of the cohort against: the trader, not the vendor, holds the kill switch.

Is the AI trading bot industry regulated?

Mostly, no, and that gap is the reason this breach matters more than it would in a regulated market.

The vendor of an AI trading bot is usually a software company, not a licensed investment firm. That means the FCA Register, the ASIC AFSL search, the CySEC list, and the SEC EDGAR database frequently return nothing for the entity you are paying. Where a bot is bundled with a regulated broker or a prop firm, the regulated piece is the broker or the funder, not the bot.

The practical rule we use: check the primary register for the entity that actually holds your money. If that entity is not listed, treat the security and data-handling claims as unverified. Start with the FCA Register for UK-facing firms, the ASIC registers for Australian ones, the CySEC list for Cyprus, SEC EDGAR and NFA BASIC for US entities, the MAS Financial Institutions Directory for Singapore, and the ESMA register for EU firms.

We do not assert a license number for any bot vendor unless we can link the register entry. If we cannot, we write "verify directly with the provider's primary regulator" and leave it there.

Table 2: security and disclosure checklist, cloud bot versus self-hosted

Dimension Cloud AI trading bot / copy trading platform Self-hosted EA or framework (Backtrader, NautilusTrader) What we score
Credential custody Broker API keys held server-side Keys stay on trader's machine Can keys be scoped and revoked in one action?
Strategy logic location Vendor servers Local Is the spec disclosed and versioned?
Third-party agent calls Common (news, sentiment, LLM endpoints) Rare unless added Are outbound calls disclosed?
Data exposed in a breach Account, positions, prompts Limited to local machine Verify with provider
Withdrawal and kill switch Vendor-dependent Trader-controlled Self-serve or support-gated?
Regulatory register entry Usually none for the bot entity None (software) Check primary register for the money-holder

Free Download: Near AI Agent Breach: Due-Diligence Checklist for AI Trading Bots
A security-focused due-diligence checklist for vetting Near AI agents and any AI trading bot before connecting broker accounts or capital.
Get Near AI Checklist

Source: Decrypt for the incident layers; checklist fields reflect our 2026 review framework. Fields not disclosed by vendors are marked verify with provider.

Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

How Zephyr AI compares

The Near AI case is a vendor-infrastructure story, so the honest comparison is about which bot you would rather be holding when the infrastructure fails. On the one concrete dimension we can score, disengagement, Zephyr AI's self-serve withdrawal and key-revocation flow beats the support-gated model we see across much of the AI trading bot market. On disclosure, its published data-handling and fee documentation is the standard we hold the cohort against, the opposite of the vague-security-page pattern the Near AI incident should make traders avoid.

Where Zephyr AI's adaptive position-sizing also edges out the reviewed cohort is in the deviation register: its outbound data flows are documented rather than inferred, which is the difference between a strategy spec you can audit and one you have to take on faith. We are not claiming any vendor is breach-proof. Hardware-backed protections have limits, as Decrypt reported. What we are claiming is that a provider which lets you cut the cord yourself, and which documents what it does with your data, is a better risk than one that does not.

The lesson from 18 exposed AI agents is not that AI trading is unsafe. It is that the security and disengagement layer deserves the same scrutiny you already give to drawdown and win rate. Put it on the checklist before you fund the account, not after.


Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.


Frequently Asked Questions

Does the Near AI breach affect retail AI trading bots directly?

Not directly. The incident involved Near AI's agent infrastructure and a flaw in the SGLang serving framework, not a specific trading bot. The relevance is architectural: many AI trading bots and algorithmic trading platforms run on the same class of hosted agent infrastructure, so the exposure pattern is transferable even though the affected systems are not.

What is SGLang and why does it matter for trading bots?

SGLang is an open-source framework used to serve and coordinate AI agents at scale. Per Decrypt, a flaw in it let an attacker reach internal systems and potentially view confidential AI requests. If a trading bot vendor relies on a similar serving layer, that layer becomes part of your risk surface whether or not the vendor mentions it.

Can a breached AI trading bot place trades on my account?

If the breach exposes a broker API key that carries trading permission, yes. That is why key scoping matters. A key that can trade but cannot withdraw, and that can be revoked in one action, limits the damage to unauthorized order flow rather than a full account drain.

Should I run an AI trading bot on a prop firm account?

Only if you understand the firm's rules on automated execution and the vendor's credential handling. Prop firm evaluation accounts often carry strict drawdown limits, and unauthorized order flow from a compromised key can breach those limits instantly. Verify the funding partner's automated trading policy directly with the firm before you connect a bot.

Does this bot work in the US under Pattern Day Trader rules?

Pattern Day Trader rules apply to margin accounts with four or more day trades in five business days and a balance under 25,000 dollars. Whether a specific AI trading bot is compatible depends on the broker and account type, so verify with the provider and the broker directly rather than assuming the bot handles PDT logic for you.

What happens if the API connection drops mid-trade?

Behavior varies by vendor and is not disclosed in the source material. The right question to ask is whether the bot holds the position until the connection returns, closes it, or leaves it unmanaged. Get that answer in writing before funding, and test it during a low-stakes session first.

Is my broker API key safe if I use a cloud bot?

It is only as safe as the vendor's credential custody. The Near AI incident shows that hosted agent infrastructure can be reached by an attacker. Ask whether keys are stored encrypted, whether they are scoped to trade-only, and whether you can rotate them without vendor support.

How do I check whether a bot vendor is regulated?

Check the primary register for the entity that holds your money, not the software brand. Use the FCA Register, ASIC registers, CySEC list, SEC EDGAR, NFA BASIC, MAS directory, or ESMA register. Most bot vendors will not appear, because they are software companies rather than licensed firms.

Can I stop an AI trading bot cleanly if the provider is compromised?

Only if the provider built a self-serve exit. The sequence is revoke the API key, close positions, and withdraw. If any of those steps requires a support ticket, your exit depends on the same infrastructure that just failed, which is exactly the scenario the Near AI breach makes worth planning for.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: Trading Industry News.

Disclaimer: Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. See our Editorial Policy.
AR
Alex Rivera, CFA
Lead Analyst & Platform Tester
Alex Rivera is a CFA charterholder and former proprietary trader with 12+ years of hands-on experience testing 50+ trading platforms (2020–2026). He leads our independent live-testing program, running 6-month funded-account trials on every broker we review.
Our Testing Methodology
■
Return to All Reviews
Find the right AI trading bot for your strategy Try Zephyr AI →