My AI Gold Explained: A Neural-Network Gold EA's Strategy and Risks

My AI Gold is a MetaTrader 5 expert advisor by Vitalii Kosov, published on 4 September 2026 and updated to version 1.82 on 21 October 2026. As of 11 November 2026 it occupies the top position on MQL5 Market's MT5 Experts Popular list, priced at 899.95 USD with a one-year rental option at 499.95 USD. It trades one symbol only — XAUUSD, gold — on the H1 timeframe, and its entries are driven by a custom neural network. The single biggest risk, stated up front: the signal logic is not disclosed anywhere on the product page, so anyone running it is trusting a black box whose decisions cannot be checked against market conditions.

How the strategy is designed to work

My AI Gold is built around one idea: gold has repeatable patterns on the hourly chart, and a neural network can learn to recognise them better than a fixed set of indicator rules. According to the product page, the system was built from the ground up using open-source machine-learning frameworks from GitHub, then optimised on live market cycles. The author describes it as pattern recognition for gold, not as a scalping system: it uses comparatively wide stop-loss and take-profit distances, targets a risk-to-reward ratio of 1:2, and is designed to hold positions longer than a typical gold scalper. Every trade carries its own fixed stop-loss and take-profit, and the author states there is no martingale, no grid and no averaging — a losing trade is closed at its stop and the next trade starts from the same base size.

Position sizing is controlled by a Risk setting that scales lot size to the account balance. Version 1.82 added a Prop Firm preset for that setting, a Fixed Lot mode, a Balance Step parameter and a Strict Lot Calculation option (off by default). A maximum-open-trades setting (1 to 99) replaced the older single/double-entry options. A Delay Deal option (the author suggests 3–10 seconds, off by default) postpones entries slightly and is described as useful for prop-firm accounts. A Wednesday close option exits positions before the triple swap charge, and an info panel shows the calculated lot size. AutoGMT detects the broker's time offset automatically, and a spread filter is mentioned in the version history.

The author recommends the H1 timeframe and gold only — the page says the EA should only be used on XAUUSD. On funding, the author gives explicit tiers: 1,000 USD is the minimum for normal risk, 500 USD or less is high risk, 200–300 USD is described as extremely aggressive, and 2,000 USD or more is the more conservative starting point. On brokers, the page says the system works with any broker and account type, while also recommending RAW Spread / ECN accounts for best results. One inconsistency is worth noting: the page recommends H1, but its "plug and play" line says to attach the EA to XAUUSD on "any timeframe".

Potential risks

1. A black box you cannot audit

This is the central risk. The product page discloses no entry rules, no indicator names, no network architecture and no training data. Users cannot tell what the model has learned, which market conditions it expects, or when it is operating outside the conditions it was trained on. Machine-learning models are fitted to historical data; when the market regime changes — a shift in the rate cycle, a change in central-bank gold buying, a geopolitical shock — patterns the model relies on can stop working without any visible warning. Because the logic is hidden, a normal losing streak and a broken model look identical from the outside, and there is no independent way to tell them apart.

2. One symbol, one timeframe

Everything depends on gold. A CPI release, an FOMC decision or a geopolitical headline can move XAUUSD sharply in minutes, and every open position is exposed to the same shock at the same time. There is no diversification across symbols or strategies. Wide stops reduce the chance of being stopped out by noise, but each stopped-out trade is a larger loss, and positions held across the close carry overnight gap risk.

3. The arithmetic of 1:2 risk-to-reward cuts both ways

A 1:2 ratio means one winning trade offsets two losing trades of the same size — but it also means losses arrive in larger units than in a scalping system. A run of five or six stopped-out trades draws the account down quickly, and recovery then requires a run of winners. The author's own deposit tiers make this concrete: at 500 USD or below the author calls the risk high, and at 200–300 USD extremely aggressive. Anyone running the system on a small account is accepting the author's own high-risk label.

4. Built for prop firms — but prop rules are strict

The author says the system is designed to pass and scale FTMO and other prop-firm challenges, and added a Prop Firm preset and a Delay Deal option. Prop-firm rules, however, usually include daily loss limits and consistency requirements that can conflict with an EA's behaviour, and failing a challenge can mean losing the challenge fee. The product page states no daily loss limit or maximum drawdown cap inside the EA itself, so compliance with a firm's rules is the user's own responsibility to verify.

5. Contradictions and gaps in the documentation

The page recommends H1 but also says any timeframe; it says the system is insensitive to broker conditions but recommends RAW/ECN accounts. The description gives no stop-loss or take-profit distances in points, no default parameter table, and no mention of a trailing stop. Users who run it on a different timeframe, a standard account or a different broker are operating outside what was actually described, and should test rather than assume.

6. Frequent updates change the system

Version 1.82 arrived about seven weeks after publication, and the changelog shows regular changes to lot calculation, entry delay, the info panel and the spread filter. Active maintenance is positive, but each update can change behaviour — a strategy that behaved one way in one version may behave differently after an update, and the user must re-validate on demo after every change. If maintenance ever stops, a model fitted to past data will age as markets evolve.

7. Popularity is not verification

The product sits at the top of the Popular list, with 8,615 free demo downloads and a 4.91 rating — but only 24 reviews and 15 activations. A Popular ranking reflects interest and downloads, not independently verified live results, and a small review sample says little about how the system behaves across brokers and market regimes. The author's price note — a rise to 999 USD after ten more copies sell — is a sales tactic, not evidence about the system.

8. Execution risks specific to gold

Gold spreads widen sharply around news and rollover, and slippage on entries and exits matters even with wide stops. The spread filter is mentioned only in a changelog fix, with no detail on how it behaves. The Wednesday-close feature exists precisely because holding gold over the triple swap is expensive — a reminder that holding costs are part of the system's economics.

Risk checklist

• Can you explain, in your own words, why the system entered each of its recent trades? If not, you are trusting a black box.
• Have you run it on a demo account, on your own broker and account type, for at least the 3–4 month evaluation window the author suggests?
• Is your deposit at or above the author's own 1,000 USD minimum for normal risk — and well clear of the 500 USD level the author calls high risk?
• Do you know what Max open trades is set to, and what happens to total exposure if it is raised above 1?
• Have you checked how it behaves around CPI, FOMC and NFP releases, and before the Wednesday triple swap?
• If you plan to use it for a prop-firm challenge, have you checked its behaviour against that firm's daily loss and consistency rules?
• Do you re-validate on demo after every version update before letting the new version trade live?
• Is the money allocated to it money you can afford to lose if gold gaps against an open position?

This article is a neutral technical breakdown for information only. It is not investment advice, and it is not a recommendation to buy or use any product.

Sources: MQL5 Market MT5 Experts Popular list and the My AI Gold product page, verified 11 November 2026.