United Airlines Bans AI Agents From Booking Flights: 3 Reasons It Matters
United Airlines Bans AI Agents From Booking Flights: 3 Reasons It Matters for AI Trading Bots
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.
United Airlines will stop personal AI agents from booking tickets on its own systems, according to reporting from The Wall Street Journal, and it will still let those agents display fares. That is a small line item for an airline's earnings. For anyone running an AI trading bot or an algorithmic trading platform against a live brokerage account, it is a much bigger signal, because it is a reminder that every automated system in finance sits on top of a permission layer that a counterparty can switch off.
We have spent the 2026 review cycle benchmarking automated retail systems, including the Ellington AI trading platform, against a field of expert advisors, crypto bots, and signal services. The pattern we keep running into is the same one United just made public. The technology works. The contract is the fragile part.
Why an airline's booking policy is a trading bot story
United's decision is the first public pushback from a major US carrier against agentic commerce, the practice of letting an AI tool transact on a customer's behalf rather than simply answer questions (investingLive). The airline gave three reasons, per The Wall Street Journal: third-party agents could expose sensitive customer information, they could complete purchases a customer never intended, and their automated actions could breach Federal Aviation Administration and Transportation Department rules about what carriers must show passengers.
Read those three reasons again and swap the nouns. A trading bot can leak account credentials or API keys. A trading bot can fire orders the account holder never intended, which is exactly what happens when a strategy deviates from its stated specification. And a trading bot can breach a rulebook, whether that is a broker's automated-trading terms, an exchange's rate limits, or a regulator's market conduct rules.
That is why we treat access policy as a first-class risk in our 2026 algorithmic testing program, on the same footing as drawdown and slippage. The technology question, can the bot place the order, is the easy one. The permission question, is the bot allowed to place the order next quarter, is the one that ends accounts.
What United actually restricted, and what it left open
United will let AI agents window-shop but not buy. Agents can display United's flights and fares, but they cannot complete transactions on the airline's website. The commercial motive is explicit: chief executive Scott Kirby said he does not want to create another middleman, echoing the years airlines spent fighting online travel agencies over fare display and sale.
The rest of the industry is split, and the split is instructive for anyone evaluating automation.
Table 1: Who lets an automated agent transact?
| Platform | Agent can browse? | Agent can transact? | Stated condition |
|---|---|---|---|
| United Airlines | Yes | No | Customer data exposure, unintended purchases, FAA and DOT compliance |
| Delta Air Lines | Yes | Yes | Authenticated assistants, extra safeguards for sensitive trips such as unaccompanied minors |
| American Airlines | Yes | Yes, for some agents | Bookings permitted through certain partners |
| Amazon | Yes | No | Sued Perplexity in November 2025 over its Comet browser agent, won a temporary court order |
Sources: investingLive, The Wall Street Journal.
The volume data tells you how early this still is. Uber reported that over Halloween week, AI agents accounted for 500 Uber Eats orders and four ride bookings, out of roughly 320 million trips. That is about 0.00016 percent of activity. The restrictions are not about today's volume. They are about who owns the customer relationship when volume arrives.
We see the same shape in retail trading. Broker API access is granted by contract, not by capability, and a bot that works flawlessly in a backtest can be locked out by a single terms-of-service update. The divergence between United and Delta is the same divergence we watch among brokers on automated order flow. One venue treats automation as a threat to its channel. Another treats it as a distribution channel with safeguards attached. Your bot's live behavior depends on which camp your broker sits in.
Do AI trading bots face the same access risk?
Yes, and in three specific ways.
First, broker and exchange APIs. Most retail brokers that expose an API attach conditions to automated order flow, and those conditions can change without a code release. We have reviewed expert advisors built for MetaTrader 4 and MetaTrader 5, crypto bots that connect through exchange APIs, and copy trading platforms that mirror positions across accounts. In every case, the binding constraint is the account agreement, not the software.
Second, prop firm and funded-account rules. Many evaluation programs restrict or prohibit fully automated trading, and a bot that passes an evaluation can be disqualified at payout stage if the firm determines the strategy was automated. If you are running an expert advisor on a funded account, the funding agreement is the rulebook, and it deserves the same care you would give a broker's terms.
Third, regulatory classification. United cited FAA and DOT rules as a reason to block agents, which is a reminder that automation can trip a compliance requirement that has nothing to do with the technology itself (FAA, US Department of Transportation). On the trading side, the equivalent rulebooks are held by the SEC and CFTC in the United States (SEC EDGAR, CFTC), the FCA in the UK (FCA Register), ASIC in Australia (ASIC Connect), and MAS in Singapore (MAS Financial Institutions Directory). We do not assert any bot provider's license status in this article. If a vendor claims to be regulated, check the claim against the primary register, including NFA BASIC for US futures firms and the ESMA registers for EU entities. If the register entry is not there, treat the claim as marketing.
Where the real risk sits for retail bot users
The United story is about access. The trading equivalent of revoked access is a forced exit at the worst possible price, and that is a drawdown event no backtest models.
Performance figures vary by strategy parameters, so consult each platform's published metrics rather than any number quoted in a review. What we can say from our review framework is that the risk metrics that matter most for a retail account are not the headline return. They are the maximum drawdown, the recovery time, and the behavior of the strategy during scheduled volatility events such as non-farm payrolls, CPI prints, and FOMC decisions. A bot that looks smooth in a backtest and then widens its stop distance during a news spike is telling you something about its true risk profile.
This is also where platform architecture starts to matter more than signal quality. A single-strategy bot is one permission away from being switched off. A platform that runs multiple strategies with shared risk limits gives you something to fall back on when one access route closes, and that structural difference is the whole point of the United analogy.
How big is the gap between backtest and live results?
Always real, and usually larger than the marketing implies.
The gap has three sources. Execution, where live fills differ from modeled fills. Access, where the live account faces limits the backtest never modeled. And specification drift, where the live strategy does something the published specification does not describe.
We treat specification drift as the most dangerous of the three, because it is invisible until it costs money. In our review framework, we compare each bot's live order behavior against its stated rules and flag every divergence. Backtest data should be verified directly with the bot provider, and any provider that will not share out-of-sample results, including the losing months, is telling you what you need to know.
Contrast this with the access question. A backtest cannot price the risk that your broker restricts automated flow, because that risk is contractual. This is the gap the United decision exposes across the whole agentic automation theme, and it is the reason we weight platform architecture, not just signal quality, in our ratings.
What to confirm before you connect a bot to a broker
Table 2: Broker and API integration, what to verify before you automate
| Integration route | Typical retail availability | What to confirm before you commit | Data status in our review |
|---|---|---|---|
| Broker REST or FIX API | Offered by many retail brokers | Whether automated order flow is permitted under your specific account agreement | Verify with the provider |
| MetaTrader 4 or 5 bridge | Common for expert advisors | Whether the bridge is sanctioned by the broker or a third party | Verify with the provider |
| Crypto exchange API | Available at major exchanges | Rate limits, key permissions, and whether withdrawal rights are attached | Verify with the provider |
| Prop firm or funded account | Restricted in many programs | Whether automation breaches the funding agreement and payout terms | Verify with the provider |
| Copy trading or social mirroring | Available on several platforms | Whether the mirror executes at your broker or on the platform's book | Verify with the provider |
Free Download: AI Agent Access Due-Diligence Checklist: What United's Booking Ban Teaches Bot Traders
A 12-point checklist to verify whether your AI trading bot's broker access, API terms, and automation permissions are sanctioned, before a United-style ban locks your strategy out mid-trade.
Get the access checklist
None of these rows carry a number in our dataset, because the answers are account-specific, and we will not invent a figure to fill a cell. The honest answer is the same one United gave its customers. Check the rulebook before you automate.
Not sure which AI trading bot fits your strategy? Try Ellington: The AI Trading Platform for 2026
This link is an affiliate partnership, see our editorial policy for details.
Can you actually switch a bot off cleanly?
Disengagement is the most underrated feature in automated trading, and it is the direct trading analogue of United's decision.
If a broker can restrict your bot, you need to know what happens to open positions when access ends. The questions we ask in our review framework are simple. Can you close all positions from the platform's own dashboard, or do you need the API that just stopped working? Does the subscription cancel at the end of the billing period or immediately? And does the platform hold your funds, or does it trade inside your own brokerage account?
A platform that trades inside your own account, under your own credentials, is structurally safer than one that holds your capital. That is not a performance claim. It is a custody claim, and it matters more than any win rate when the access layer fails.
How Ellington Compares
Here is where the United analogy pays off for a retail trader choosing a platform.
The reviewed category, single-strategy AI trading bots and signal providers, shares United's problem in reverse. Each one depends on a narrow access route: one broker connection, one exchange, one strategy, one permission. When that route closes, the strategy stops. Ellington's multi-strategy automation spreads execution across multiple strategies under a shared portfolio-level risk framework, which is the concrete dimension where it wins against the single-strategy field. If one strategy or one access route is restricted, the platform has somewhere else to run, rather than a single point of failure.
That is the same lesson United just taught the agentic commerce industry, applied to your account. Concentration in one permission is a risk, whether you are an airline protecting its booking channel or a retail trader protecting a funded account.
What this means for your portfolio in 2026
The headline is about travel. The takeaway is about control.
AI developers are betting that agents will become a new way to buy things. OpenAI bought travel-booking startup Odessia, and Meta says travel planning is a top use for its Muse agent (investingLive). Those bets assume open access. United, Amazon, and the courts are testing that assumption in real time. Amazon sued Perplexity in November 2025 to stop its Comet browser agent from shopping on Amazon's site, and later won a court order that temporarily blocked it.
The timing lands in the same window as Nvidia's results and Anthropic's market debut, both flagged as tests of how much AI revenue current valuations assume. If restrictions spread, the pattern the source article describes is that merchants with strong direct channels protect their data, their pricing power, and their advertising reach, while AI platforms get pushed toward partnership deals rather than open access. Retail trading has a version of that same dynamic. Brokers that own the customer relationship will decide which automated systems get to trade, on what terms, and at what cost.
For a retail trader, the practical response is not to avoid automation. It is to avoid concentration. Diversify across strategies, keep custody of your own capital, and treat access as a risk you manage, not a feature you assume.
Not sure which AI trading bot fits your strategy? Explore Ellington, the AI trading platform built for 2026
This link is an affiliate partnership, see our editorial policy for details.
Try Ellington: The AI Trading Platform for 2026
Try Ellington: The AI Trading Platform 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 an AI trading bot work in the US under Pattern Day Trader rules?
Pattern Day Trader rules apply to the account holder, not the software, and most brokers apply the same designation to automated order flow as to manual orders. Confirm the current thresholds and your account type with your broker before you automate a high-frequency strategy in a small account.
Can I run an AI trading bot on a prop firm account?
Sometimes, but many funded-account programs restrict or prohibit full automation, and a bot that passes an evaluation can be disqualified at payout stage. Read the funding agreement before you connect anything. If the agreement is silent on automation, ask the firm in writing.
What happens if the broker API connection drops mid-trade?
It depends on where the strategy lives. If the bot runs on your machine and only sends orders, a dropped connection can leave positions open with no management. If the bot runs on the platform's servers, the platform may keep managing the trade. Either way, know the answer before you fund the account, and keep a manual exit plan.
Is United's ban a sign that AI trading bots will be restricted too?
Not directly, because United's decision concerns airline bookings, not securities. But the underlying logic, that the platform owner controls whether an agent can transact, applies to broker APIs and exchange connections. Access to automation is a commercial decision, not a technical one.
How do I check whether a bot provider is actually regulated?
Go to the primary register for the relevant jurisdiction: the FCA Register in the UK, ASIC Connect in Australia, SEC EDGAR or NFA BASIC in the United States, or the MAS Financial Institutions Directory in Singapore. If the vendor's claim does not match a register entry, treat it as marketing.
Why does backtest performance differ from live results?
Three reasons: execution differences between modeled and live fills, access limits the backtest never modeled, and specification drift where the live strategy departs from its published rules. Backtest data should be verified directly with the provider, including out-of-sample and losing periods.
Can I stop an AI trading bot cleanly?
Usually, but the details matter. Check whether you can close all positions from the platform's own dashboard without the API, whether the subscription cancels immediately or at period end, and whether the platform holds your funds or trades inside your own brokerage account. A platform that trades inside your own account is structurally safer on the disengagement question.
Do AI trading bots work with crypto exchanges as well as brokers?
Many do, and crypto exchange APIs are often more permissive than broker APIs. That flexibility carries its own risks, including rate limits, key permissions, and the possibility that withdrawal rights are attached to the same key the bot uses. Scope the key permissions carefully.
What is the single biggest risk in automated trading right now?
In our view it is concentration in a single access route. A bot that depends on one broker, one exchange, one strategy, and one permission has a single point of failure that no backtest prices. Spreading execution across multiple strategies under shared risk limits is the structural answer.
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.