My AI Gold EA Review (2026): A '+1000%' Gold EA Whose Headline Is a 300 Dollar Account and Whose Calmer Signal Has 28 Trades

44.0 / 100

weighted verdict score on eight published criteria

PF 1.97 and 2.79

profit factor of the two linked signals, on 116 and 28 trades, with an average loss near twice the average win

$899.95

Market price on 29 Oct 2026, or $499.95 for a year's rental, with a free demo; the price rose from $799.95 in about two weeks

Is My AI Gold legit? The verdict at a glance

My AI Gold is a real product, not a hollow listing. It has a long description, six set files, a user guide in many languages, a seller who answers comments, seven updates in eight weeks, 14 five-star ratings and two public signals. We found no allegation of non-payment, no deleted criticism and no loss complaint aimed at the vendor. It scores low for three reasons. The record is short: 24 weeks on one signal and seven on the other, against the 12 months our anchors ask for. The marketing claims several things the numbers do not show, from the size of the live track record to the risk-to-reward ratio. And the default setting is, in the seller's own words, high risk. The scorecard and radar show how the total was built.

What My AI Gold actually is

My AI Gold is an Expert Advisor for MetaTrader 5, sold on the MQL5 Market by Vitalii Kosov of Russia, who describes himself as CEO at 'Beast FX' AI Trading Experts. His seller profile shows one product and three signals. The profile also shows a rating figure of '3 (139)' that we could not interpret with confidence, so we did not use it. The product was published on 4 September 2026, showed version 1.82 last updated on 21 October 2026, and carries 15 activations, 5,685 free-demo downloads, 40 comments and a rating of 5.0 from 14 reviews. The page says '155 purchases per month'.

The pitch is an 'AI-driven gold trading revolution'. The description says it is built on 'custom neural networks' using 'advanced machine learning models aggregated from elite open-source GitHub frameworks', and that it is 'optimized specifically for high-precision gold pattern recognition'. It then lists what it is not: not a fragile scalper, and 'absolutely NO Martingale, NO Grid, and NO Averaging', with strict stop loss and take profit on every trade. It claims a 'healthy' 1:2 risk-to-reward ratio, says it is 'prop firm ready' and 'broker independent', and tells you to attach it to 'any timeframe' on XAUUSD, although a later section says it is designed for H1. It also warns that not every month will be green and that the EA should be judged over at least three to four months.

What the page does not contain matters as much. It describes no architecture, no inputs, no training data and no retraining schedule, and shows no backtest figures. The only performance evidence is the pair of signals linked in the first line. Everything below tests the claims against those signals, against an independent Strategy Tester write-up of version 1.82, and against what buyers said in the comments.

Our listing against the Market product: the exact differences

Because we sell a listing under this name, the differences are part of the verdict. A buyer comparing the two should see them side by side. Everything in the left column was read on the Market page and everything in the right column on our own listing, both on 29 October 2026.

How we judged it, and who has something to sell

Evidence on EAs is never neutral. Here is every party near this product and what each stands to gain, ours included.

• The developer sells the EA, sells a signal subscription (777 USD a month on each of the two signals, with no subscribers shown on either) and runs the discussion group. His record is public, which counts for something, and it is also his sales page.
• MQL5 and MetaQuotes host the Market, the reviews and the signals. The signal statistics and every platform notice quoted below are the platform's own.
• AlgoToolbox, whose tester ran version 1.82 in the Strategy Tester, discloses a broker affiliate link on its page. We use its mechanical findings and its numbers, and treat its opinions as opinions.
• Buyers who wrote reviews include three readable ones, all positive. Testimony, not data, and none can be checked from outside.
• ForexCopilotEA sells a listing for this EA. We gain if you buy it, which is why this verdict opens with that fact and applies the same rules as every other.

My AI Gold live track record: the +1000% claim, number by number

The first line of the description reads 'LIVE TRACK RECORD +1000%' and links MQL5 signal 2389616. That signal is named 'My AI Gold EA High Risk', runs on the Weltrade-Live server at 1:500 leverage, and is a MetaTrader 4 account, although the product it advertises is for MetaTrader 5. It started with 300 USD, with no deposits and no withdrawals. A platform-computed signal is the best public evidence a vendor can show. It still has limits: the vendor runs the account, the broker is one broker, the base is tiny and the history has oddities we list below. Here is what it shows on 29 October 2026.

The signal says +1,828%, not +1000%, so the listing understates the headline. But the percentage is a statement about the base. A 300 USD account that grows to 5,784 USD has compounded: the lot follows the balance, so a good run of trades feeds itself. The profit in dollars is 5,484.26 USD. The same arithmetic applied to a 10,000 USD account that makes 14.66% gives 1,466 USD, and that is the honest comparison between the two signals. Percent figures on tiny accounts are the standard way a track record looks larger than it is, and the listing itself warns that a 200 to 300 USD deposit means 'extremely aggressive trading, with risk levels close to a casino-style account gamble'. The headline account is exactly that account.

The payoff is the second thing to read. A 79.31% win rate with an average winner of 120.87 USD and an average loser of 234.82 USD gives 0.7931 times 120.87 minus 0.2069 times 234.82, which is 95.86 minus 48.58, or 47.28 USD per trade, the platform's figure. The loser is 1.94 times the winner. The break-even win rate for those averages is 234.82 divided by 355.69, about 66%. The signal wins 13 points more often than that, which is its entire edge. If the win rate fell to 66% on the same averages, expectancy would be zero. The listing's claim that you 'do not need an unrealistic, sky-high winrate' is therefore not what the data shows: this system needs a high win rate, and so far it has one.

The three drawdown figures differ because they are measured differently. Maximal drawdown by balance is the largest fall from a peak in dollars, here 894.78 USD, or 18.84% of the peak at that time. Relative drawdown by balance is the largest percentage fall, and it is 43.59%, which happened early when the balance was small. Relative drawdown by equity counts open positions, and is 12.53%. Every one of them describes a real account, and the largest is the one a new buyer on a small account is most likely to meet. A system that can fall 44% of its balance in a run of losses is a High risk system whatever its stop losses say.

The calmer record: My AI Gold on a 10,000 dollar IC Markets account

The second signal the listing links, 2390451, is named 'My AI Gold ICmarkets 10k Low Risk'. It runs on MetaTrader 5 at ICMarketsSC-MT5-3, leverage 1:500, from 8 September 2026, with a 10,000 USD start, no deposits and no withdrawals. It is the closer match to the product on sale. It is also seven weeks old, and the platform's own current notice reads that the number of deals is too small to evaluate trading quality and that the results of a newly opened account may be of random nature.

Read it for its shape, not its size. Four losing trades of about 205 USD each against 24 wins of about 95 USD is the same negative-skew pattern as the first signal: the loser is 2.15 times the winner and the break-even win rate is 205.09 divided by 300.37, about 68%. The signal wins 86% of the time, 18 points above that. The worst trade of 212.88 USD is 2.1% of the account, and the maximum deposit load of 0.90% shows the lot is small. That is what 'low risk' means here: a small lot on a large account, not a different strategy. The same design at a lot sized for 300 USD produced the High Risk signal.

Twenty-eight trades cannot confirm a win rate of 86%. If the true rate were 70%, a run of 24 wins in 28 would be unlikely but not impossible, and the confidence interval on such a sample runs from the high sixties to the mid nineties. What the signal does show is that nothing broke in seven weeks and that fills at one IC Markets server were good enough to profit. The copy-quality table is a warning on the other side: MQL5 lists the slippage of subscriber servers on this signal, and it ranges from 1.00 on ICMarketsSC-MT5-4 and 1.43 on ICMarketsEU-MT5-5 to 8.32 on Tickmill-Live, 9.89 on RoboForex-Pro, 14.50 on GlobalPrime-Trade, 15.00 on Exness-MT5Real50 and 21.16 on ICMarketsSC-MT5-3, the signal's own server, over 114 deals. The units are the platform's. The point is the spread: the same orders cost very different amounts to execute on different brokers.

My AI Gold claims against the evidence

Most of the score comes from the distance between what the page says and what we could check. The table sets the main claims against the sources, and says plainly where we could not decide.

My AI Gold backtest: what an independent Strategy Tester run found

The listing publishes no backtest. The best public one is by AlgoToolbox, updated on 27 October 2026, run by the tester Eko Martha Kusumah on version 1.82 using the free Market demo. The settings were XAUUSD H1 from 3 December 2020 to 21 October 2026 on an Exness raw-spread account with a 2,000 USD deposit and 1:500 leverage, in real-tick mode, but with only 4% to 7% of the history at real-tick quality and the rest generated. That is a real limitation: generated ticks flatter or distort the fills of a system that trails stops and trades the Asian and London sessions. The page discloses a broker affiliate link.

Three findings deserve attention. First, the conservative set is a reasonable-looking system on paper: 72% winners, a profit factor of 1.76, 11% equity drawdown, 71 months of history. But its edge is a win rate, not a payoff. An average loss of 63.48 USD against an average win of 43.34 USD means the break-even win rate is 59.4%, and the run wins 72.14%, a margin of 13 points that mirrors the live signals. Second, the aggressive set shows what the 'very aggressive' tier means: a 90.24% loss of balance and 477 days to recover, at about 26% risk per trade. The tester also reports a final balance of 49.6 million USD for it, which came from compounding to 200-lot positions that no broker would fill, so it is meaningless as a forecast and a good illustration of why percent returns on compounded lots mislead. Third, the post-launch figures are the most honest part, because the settings were re-optimised in version 1.51 on 24 September 2026, so most of the six backtest years are in-sample. The out-of-sample window is a few weeks: +467.89 USD on 26 trades with a profit factor of 1.25 for the conservative set, and +10.54 USD on 24 trades for the prop set. Both are positive and both are too small to carry weight.

For how a tester run differs from live trading, see our guide to backtest versus live results and our guide to reading an MT5 backtest report. For the risk of results that were tuned on the data they are shown on, see our guide to optimising an EA without overfitting.

How My AI Gold trades: what the tester shows about the mechanics

The most useful part of the AlgoToolbox write-up is the plain description of behaviour. According to the tester, the EA uses market orders only, holds a maximum of two positions (one in the prop set), and trades inside a window from 15:00 to 11:00 with automatic GMT handling. The stop loss is fixed at about $40 and the take profit at about $25 (inputs TKProfit 250 and RST 400). A trailing stop begins at about $16 of profit (TrailStart 160, TrailStep 5). Friday close is on, the Wednesday close before triple swap is off by default, the news filter is off and the Recovery input is off. The lot follows the balance, at roughly 1.8% risk per trade on the conservative set. Entries cluster in two bands of server time, 01:00 to 09:00 and 12:00 to 20:00.

Put together, this is a rule-based intraday system on H1 that enters in particular windows, risks a larger stop than its target, lets the trailing stop turn some trades into small wins, and cuts risk by closing at the Friday session end. The tester counted 15 trades that reached the take profit, 766 exits tagged as stops (a category that includes trailing stops) and 242 positions closed on Fridays at 20:00 server time. Those three groups sum to the 1,023 trades. It means a 72% win rate is largely produced by the trailing stop locking in small gains, not by a target being reached, which is consistent with an average win of 43 USD against a take-profit distance of about $25 per unit that the lot multiplies. It also means the strategy's edge depends on the path price takes after entry, and on how your broker fills and trails stops.

Is that a 'neural network'? We do not know, and the tester notes that behaviour looks rule-based. A model can produce entry signals while the exits are fixed rules, and many commercial products do exactly that. The listing says the models were 'aggregated from elite open-source GitHub frameworks' and does not say whether anything is trained, retrained or updated on a live account. The honest position is that the claim is unverifiable. It does not need to be true for the EA to work, and it does not follow that the EA works because the claim is true. Judge the product on the signals and the tester, which describe behaviour, and not on the label.

Use the calculator with the figures above. Enter a win rate of 79%, an average win of 121 and an average loss of 235 and you reproduce the High Risk signal's expectancy of about 47 per trade. Then lower the win rate to 70%: expectancy falls to about minus 14 per trade, and the edge is gone. That is how sensitive a payoff of 0.5 is to the win rate, and it is why a young record of 116 trades cannot settle the matter.

My AI Gold drawdown and risk: a default that is set to high

My AI Gold is better behaved than a grid. Every trade has a stop, nothing is added to a loser, the tester saw at most two positions and the worst trade on the larger signal was minus 522.91 USD on a balance that was then far above 300 USD. That would keep the label at Medium if the story ended there. It does not, for the reasons below, and we label the product High.

• The default is high risk. On 30 September the seller answered a buyer that 'the default settings use a HIGH risk level for demonstration purposes' and advised larger accounts to choose a more conservative set file. A buyer who attaches the EA and presses go is running the setting the seller calls high.
• Drawdown scales with risk. Signal B's maximal balance drawdown is 3.41%; signal A's relative balance drawdown is 43.59%; the tester's aggressive set fell 90.24%. The same code spans three orders of outcome, so the set file you choose matters more than the product.
• Negative skew. The average loss is about twice the average win on both signals and in the tester's set (1.46 times). A cluster of three or four losers erases most of a good week.
• Broker sensitivity. A buyer reports that live trades on the signal account are absent from his own and from his own backtest. He reads that as the EA being sensitive to the broker, and it matches the slippage spread above. Your broker's results can differ from the signal's.
• A user-error loss on record. The same reviewer says he lost $4.6k through a risk setting he got wrong, and was later up $2.8k. He attributes it to his own mistake, and we report it as that: it shows how large the cost of a wrong setting can be.
• The aggressive and high-risk sets are described by the seller himself as 'close to a casino-style account gamble' for deposits of $200 to $300.

What buyers say about My AI Gold

The listing shows a rating of 5.0 from 14 reviews, with every rating five stars, in a product less than two months old. We could read three review texts; the rest load by script. One, from theopott on 24 October 2026, says 'very satisfied with this EA, and the developer is responsive'. One, from Adel Mohamed on 20 October, says 'everything is perfect, so far'. The third, from Pablo Sanz Moniche on 19 October, is the substantial one. He says it is difficult to find EAs that match their backtests and that this one does; that some trades on the live signal are not mirrored in his own account, but that the same trades are absent from a backtest of that period; that the EA is 'highly sensitive to each broker's specific conditions, which isn't a bad thing'; and that his only warning is to double-check risk levels, since a wrong setting cost him $4.6k, after which he was 'currently up $2.8k'. He adds that he would be up 10% in a month with conservative risk.

Two cautions apply. A rating of five stars from 14 buyers in eight weeks is encouraging and says little about performance in a bad month, and the three texts include one that describes a large loss. And the comment section, 40 comments strong, is largely requests for set files, instructions and prop-firm advice, which suggests the defaults are not meant to be used as they come. We found no negative review, no loss allegation against the vendor and no reply of his under a review. In the comments he answers requests, repeats his recommendations, promised on 30 September to add a clearer status panel in the next update, and on 10 October posted a changelog for version 1.71: a panel that scales to the screen resolution, a Wednesday-closing-time input before triple swap, minor bug fixes and a slightly improved entry logic, noting the update 'is not mandatory'.

My AI Gold price: $899.95, a rising ladder and a one-year rental

On 29 October 2026 the Market price was $899.95 to buy or $499.95 to rent for one year, with a free demo. The page says that when the next ten copies are sold the price will permanently rise to $999. Two sources quoted $799.95 earlier in the month, a blog on 12 October and the AlgoToolbox review on 25 October, so the price rose by $100 within about two weeks. The same page shows 155 purchases a month, which makes ten copies a matter of days, not a scarce supply. We read the line as urgency pricing, and the trust score reflects it.

Set the price against the evidence. A buyer at $899.95 with a 2,000 USD account pays 45% of the account to the vendor before the first trade. At the tester's conservative-set result, 13,793 USD over nearly six years, the price is recovered, but that figure is an in-sample backtest on mostly generated ticks. At the larger live signal's 47.28 USD per trade it would take about 19 trades, but that signal is the 300 USD high-risk account. The honest range is wide, and the rental at $499.95 for a year is the cheaper way to find out where in the range a given broker and a given set file land. The demo, with 5,685 downloads, runs in the Strategy Tester only, so it shows you the backtest and not the live behaviour.

Remember the other costs: a raw or ECN account, a Windows VPS, which the seller recommends, and the time to choose and test a set file. Our own listing prices an EA-only edition at $26.99 and asks you to decide whether to buy a copy of version 1.82 without the manual, without the five extra set files and without the author's support. Read what is included, read the 'all sales final, no refunds' line, and weigh the provenance question we raised above.

How to run My AI Gold without fooling yourself

A product with six presets, a high-risk default and results that differ by broker rewards a short, disciplined routine. First, decide your risk budget in dollars before you open the set files: the amount you can lose in a week, which should be at least three average losses of the lot you intend to use. Second, pick the set file whose stated deposit matches yours, the low-risk file from about $2,000 and the normal-risk file from about $1,000, and do not move up a tier because the signal on the page looks good. Third, use H1 on XAUUSD, whatever the 'any timeframe' line says.

Fourth, check your broker. The seller recommends a raw or ECN account, says the EA is 'not highly sensitive' to spreads because its stops are wide, and still the copy-quality table above shows slippage ranging from 1 to 21 on the very same orders. Run the demo on your broker's data, then run the EA on a demo or cent account for at least the three to four months the seller himself recommends for judging performance, and compare your fills to the signal's, trade by trade. Fifth, set the Wednesday triple-swap close deliberately, since the tester found it off by default, and decide about the news filter, which is also off. Sixth, keep one set file for a long time: a buyer who changes set every fortnight never builds a record of their own and cannot tell a bad set from a bad month.

Seventh, treat the signal as a sales page, not a promise. The High Risk signal's monthly growth of 29.61% and annual forecast of 359.27% are the platform's extrapolation of a short run on a tiny base, and the platform flagged 'too much growth' on it twice. A forecast of that size is a reason for caution, not a plan. If a conservative profile on 10,000 USD returns about 5% a month, as the second signal shows in seven weeks, that is a good result and an unproven one, and you should size your own expectations to what you have measured on your own account.

The score, criterion by criterion: 44.0 out of 100

The total is 44.0 out of 100. The grade A on evidence is earned by two platform-built signals; the score stays at 4 because the longer one is on a different platform from the product, carries date inconsistencies and is run by the vendor, and the cleaner one has 28 trades. Risk, at 4, gives credit for stops on every trade and no grid or martingale, and takes it back for a high-risk default and a 43.59% relative drawdown on the headline account. Profitability, at 5, is the strongest heavy criterion: positive expectancy on two signals and a tester, held back by a payoff near 0.5 and the absence of 12 months. Trust, at 3.5, is the lowest score, because the sales claims are further from the data than the product is. Support, at 7, is the highest.

Who My AI Gold fits, and who it does not

• It fits a trader with at least a couple of thousand dollars who wants a rule-based gold system on H1, will pick the low or normal-risk set file on purpose, and is prepared to run a demo for several months before sizing up.
• It does not fit anyone with a deposit of a few hundred dollars who is attracted by the +1,828% headline. That is the casino-style case the seller warns about, and the 43.59% relative drawdown was measured on exactly that account.
• It is risky for prop-firm challenges unless the prop set is used at a small lot and the daily limit is respected: a payoff of 0.5 can breach a daily limit in a cluster of two or three losers, and no audited challenge result is linked.
• It is a poor choice for a buyer who is paying mainly because only ten copies remain at the current price, or who expects every month to be green. The seller says as much himself.

How My AI Gold compares with other gold EAs we have scored

Its closest relative in this series is Gold Beaver: a gold EA with stops on every trade, a high win rate and a payoff well below one, young signals and an urgency price. My AI Gold's record is a little longer and larger, its price twice as high, and its default setting is the more aggressive one. Pulse Engine is the opposite case: a large live sample at a profit factor near 1, with far smaller losses per trade. Luna AI Pro is another product that leans on AI wording, and the same rule applies to both: we score the behaviour and not the label. Its 44.0 reflects a promising and young record under marketing that runs ahead of it. For how these system types differ, see our strategy comparison and our gold EA category.

Questions to put to the developer before you pay

1. Why does the High Risk signal show a start date of 3 September 2026 but 24 weeks and 76 trading days, and why is it a MetaTrader 4 account for a MetaTrader 5 product?
2. Which set file, risk level and version does each signal run, and has either changed since it started?
3. What is the largest loss and the longest losing streak you have seen on any customer account, at what lot and which set?
4. Where does the 'healthy 1:2 risk-to-reward' come from, given a fixed stop larger than the target and an average loss twice the average win?
5. Which part of the system, if any, is a trained model, and is anything retrained after release?
6. Why is the default a high-risk setting, and would you ship a conservative default?
7. Can you publish an audited prop-firm challenge result and a real-tick tester report for the conservative set?
8. How does the price ladder work, what happens to existing owners, and what happens to a rental that ends?