What is the biggest risk with AI-generated trading bots? - page 2

[Himaya]

The risk I would add is that AI-generated logic is usually not obviously broken. It is plausible and mildly positive, which is more dangerous than being wrong in a visible way.

I walk-forward tested a simple EMA crossover EA recently (21/55, EURUSD H1, ATR stops, fixed percent risk, 10,000 USD deposit):

In-sample 2023-2024: 10,000 -> 10,638 (+6.4%)

Out-of-sample 2025-2026: 10,000 -> 10,084 (+0.84%)

A quick look says "small but positive, keep it". The number that actually kills it is different: 413 trades for 84 USD net, which is about 0.20 USD per trade. That is below spread and commission on any real account. So the strategy has no edge, it just has no obvious failure either.

So alongside overfitting, a practical check I would suggest: divide net profit by number of trades and compare it to your realistic round-trip cost. If per-trade profit is under your costs, the backtest is noise no matter how the equity curve looks.

One more thing that surprised me. The risk management code can be completely sound while the strategy underneath is worthless. In the same test the daily loss limit and the drawdown limit were never breached once across 413 trades. They worked exactly as written, they just had nothing worth protecting. People see working risk controls and assume the logic behind them is validated, and those are two separate questions.

[Alex Holloway]

David Chidiebere Chinweike:

What is the biggest risk with AI-generated trading bots?

Bad logic, poor risk management, overfitting, or users not understanding the code?

Th biggest risk is blind trust. People run AI generated code without really understanding it or taking the time to test it properly. That's how bad logic, poor risk management and overfitting get overlooked

Someone asks AI to build a profitable EA, sees that it compiles and looks good in a backtest, and suddenly thinks they've got a money printer. Meanwhile, they haven’t checked the lot sizes or whether that "maximum drawdown" setting actually does what they think it does

Before putting money behind it, understand why it opens trades, how much exposure it takes on and what happens when things go wrong. Test it on data you didn't use for optimization. A beautiful backtest alone wouldn't convince me to trust it with my account

AI can help you write the code. But you're still responsible for understanding and validating what you're putting your money behind. If your entire risk management plan is "the AI said it works," you're paying for that assumption with your own money. And the market doesn't give a damn who wrote your code.