AICaverDo AI Trading Bots Actually Work?
AI trading bots can analyze markets and place trades, but evidence for a durable edge after fees, slippage and risk remains weak.
Yes — if “work” means analyzing markets, generating signals, and executing trades automatically. But there is still no convincing evidence that today’s AI trading bots can consistently beat the market after fees, slippage, and risk.
The key is to separate three different claims:
- the bot can trade;
- the bot can make money for a while;
- the bot has a durable market edge.
Those are not the same thing.
In a 2026 experiment, six AI models received $10,000 each and traded prediction markets autonomously. On Kalshi, they made 2,916 trades over 57 days. All six finished in the red.
The most striking result came from Grok. It correctly predicted 71.4% of the contracts it held to settlement — and still lost money.
That sounds paradoxical only if accuracy is treated as profitability.
A strategy can win seven trades out of ten and still lose overall if the losing trades are much larger than the winners. Fees, spreads, slippage, position sizing and drawdowns matter just as much as prediction accuracy.
| Result | What it actually proves |
|---|---|
| High prediction accuracy | The model is often directionally right |
| Strong backtest | The strategy worked on historical data |
| Short-term profit | It made money in one market period |
| Profit after costs, risk and a benchmark | Evidence of a real trading edge |
Backtests are especially tricky for language models because they may have encountered parts of market history during training. A model can appear to “predict” the past partly because the past is not genuinely unknown to it.
That does not make AI useless for trading. AI is already valuable for reading financial documents, analyzing news, comparing assets, generating hypotheses and preparing trades.
The strongest results so far tend to come from hybrid systems in which AI handles information and analysis while separate mechanisms control portfolio construction, risk and execution.
AI Caver take
AI looks much more convincing today as a component of a trading system than as an autonomous market oracle.
Read the full research
The AI Trader Was Right 71% of the Time. It Still Lost Money