Can AI Pass a Prop Firm Challenge? The Direct Answer, and the Three Things the Question Hides

Yes. Automated software passes prop firm evaluations, and it is common enough to be unremarkable. Prop firms commonly permit expert advisors on at least some account types, though permission is structured differently by firm, by account and by platform, and firms revise it, so read the rulebook on your own firm's site before you automate anything. So the short answer to can AI pass a prop firm challenge is yes, with that condition attached. The longer answer matters more, because the question carries three assumptions with it, and each one leads people to buy the wrong product.

The first assumption is that AI means a learning model. In this market it usually does not. The second is that passing is the hard part. It is the easy part. The third is that a chat model could do the job with a good enough prompt. It cannot, because it never touches your account.

One disclosure first, since it changes how you should read the rest. This blog is published by PraxAI, and we sell trading automation to prop firm traders, which is a direct conflict of interest on this exact question. Where we cite our own numbers further down, we say where they are published so you can open the account and check the curve yourself. For the wider survey of the category, start with our guide to [AI trading bots for prop firms](/blog/best-ai-trading-bot-prop-firms-2026).

Open the panel on most software sold under this label and you find if then statements. If the daily loss reaches the number you set, close the position and stop trading. If spread is wider than this threshold, do not enter. If a high impact event falls inside the window, pause. Nothing there is learning. It is a decision a human wrote down in advance and the machine executes without discretion.

That is worth saying plainly rather than as an accusation, because deterministic rules are not a downgrade. For the part of the job that decides whether you pass, they are the correct design. A loss cap that behaves the same way every single time is a better cap than one that weighs the situation and decides whether the limit applies today. Determinism also makes any claim checkable: run the same period twice and compare the trade lists line by line.

The problem is narrower than the label: paying a premium for two letters that do not describe the product, and being unable to get a straight answer about which part learns and which part is a rule. The test for that is set out in [how to tell whether a trading bot is actually AI](/blog/is-any-trading-bot-actually-ai). Ask any vendor to name the parts of their own software that are not AI. An honest one will do it in a sentence.

An evaluation is a closed problem with published constraints. There is a target, a set of limits, usually a minimum number of trading days, and a defined finish. Any system that follows a risk policy without negotiating can work inside a specification like that, which is why the answer to the headline question is yes.

A funded account is an open problem. There is no finish line, so the only way to end it is to break a rule or to give the profit back. This is where the number that actually describes the industry lives: only 1 to 3 percent of funded traders keep the account long term. That figure, not a pass rate, is the one to hold in your head when you decide what to buy. Repeat challenge fees can add up to 2,400 dollars a year for a trader who keeps restarting.

The failure modes after funding differ from the ones during an evaluation, and they are covered in [why funded traders lose the account](/blog/why-funded-traders-lose-the-account). If you are buying software to reach a payout rather than to clear a fee, read [how the first payout timeline actually works](/blog/first-payout-timeline) before you buy. A product optimised to hit a target inside a window is not automatically one that survives the year after it.

No, and the reason is structural rather than a matter of capability. A general purpose language model has no connection to your broker. It does not place orders, does not see your equity change tick by tick, does not know how many dollars you have left before today's loss limit, and is not running while you sleep. It answers a question at the moment you ask it, using the text you paste in.

That makes it a reasonable tool for reading a rulebook, explaining what a trailing drawdown does, or checking arithmetic. It is not automation. Asking a chat model for a strategy gives you a description, and a description does not size a position, attach a stop, or shut the account down when a limit approaches. Every part of a challenge that is actually hard happens in the gap between having a strategy and executing it consistently for weeks.

There is a second problem underneath the first. A model producing plausible text is not accountable for the number it produces. If it suggests a lot size that leaves you 0.4 percent from a daily limit, nothing in the system stops that order. The stop has to exist as code sitting between the decision and the broker, with your firm's limits set inside it.

Strip out the marketing and approval comes down to four things. All four are constraint problems, which is the reason software does well at them and the reason the reasonable version of the answer is yes.

Every one of these is commonly structured differently by firm, by account type and by platform, and firms revise them. Nothing here states any firm's current rules. Open your own firm's rulebook, write the four numbers down, and screenshot the page with the date on it.

The honest case for software here is narrow and it is behavioural. Code does not feel fear at the fourth losing trade, does not revenge trade after a bad morning, does not widen a stop because the position is almost back, and does not decide that today is an exception. It also does the arithmetic a human will not do at speed: measure the distance to the daily limit before every entry, and refuse the order if taking it would put the account inside the buffer.

The list of things it does not fix is just as short. A rule change at your firm can invalidate a configuration overnight, which is why an update policy matters and why [what happens when prop firms change the rules](/blog/when-prop-firms-change-the-rules) is worth reading before you commit. Execution is real: slippage, spread widening and a position moving during the close all sit between the intended stop and the filled one, which is why an internal limit should sit below the published one rather than on it. And no configuration removes market risk. A system can follow every rule perfectly and still have a losing month.

One area needs a specific warning. On futures, whether you may run automation at all depends entirely on each individual firm's policy. We will not tell you that bots or expert advisors can run at Apex, Topstep, Tradeify or any other named futures firm, because that is the firm's decision and it changes. Ask support directly and get the answer in writing before you pay for anything.

Since anyone can claim their software passes challenges, the only version of the claim worth anything is one you can open and inspect. Here is ours, stated as what we publish rather than as an independent audit, because it is not one.

PraxAI publishes 24 approved challenge accounts at praxai.io/results. Each one shows the interactive equity curve and, next to the gain, the real maximum drawdown that produced it. The firms include FTMO, The5ers, E8 Markets, The Funded Trader, FundedNext, Instant Funding, Alpha Capital, FunderPro and FundingPips. Named examples from that page: an FTMO 100K at plus 10.28 percent with 1.96 percent maximum drawdown over 12 days, an E8 Markets 200K at plus 8.31 percent with 1.73 percent over 14 days, and an Instant Funding 100K at plus 10.33 percent with 1.19 percent over 9 days.

The number we would point at is not any of the gains. It is that the highest maximum drawdown across all 24 accounts is 3.35 percent, because that describes how much room was left when the target was reached. Those figures are published by PraxAI on our own results page, they are our own accounts rather than a third party audit, and you can open each one and read the curve there. Nothing about them is a promise that any particular challenge will be approved.

If you take one thing from this post, take the question to ask next. Not can AI pass a prop firm challenge, because the answer is yes and it tells you nothing about which product to buy. Ask instead: show me a passed account with the maximum drawdown displayed next to the gain, and tell me which parts of the risk control are fixed rules I can set myself rather than a model I am asked to trust.

A vendor who can answer both has given you something checkable. One who answers with a pass rate, a total payout figure or a promise that you will be approved has given you a marketing number, and no honest seller can promise approval, ourselves included.

The configuration we validated on gold takes one position at a time, with no martingale and no grid, and a fixed stop loss. PraxAI GUARD holds the daily loss and drawdown limits as code level locks that you set, and it shuts the system down before the firm's limit rather than at it. That is deliberately not AI, because for a hard limit you want predictable ahead of clever. None of that makes an account safe. Software can carry the discipline. It cannot carry the risk.