Do AI trading bots make money? What the public record shows

Published 13 October 2026

In September 2026 we looked for any trading bot, AI or not, with a verified record of making money after costs. We found plenty of verified profits, almost all at firms with an edge a retail account cannot copy, and not one audited track record for a retail bot.

How this was made: the code, the research runs and the first drafts of this text were produced by an AI assistant (Claude) working under the operator's direction; the operator made every decision and checked the numbers. All trading was on paper: real prices, no real money. Nothing here is investment advice.

Where verified profits are

Automated trading clearly makes money for some. Market makers and high-frequency firms publish audited results: Virtu reported net income of $912.3 million for 2025 and Optiver €1.769 billion. Their edge is structure: they are paid for providing prices, they trade in microseconds, and they win on a sliver of each of billions of trades. Large quantitative and multi-strategy hedge funds reported gains of roughly 10% to 33% in 2025, against 17.7% for the S&P 500, while trend-following indexes made about 2% to 3%.

What we could find about retail bots

No retail bot platform or copy-trading service we checked publishes audited results across its users. Figures like "93% of copiers made money" are marketing.
On Hyperliquid, where vault results are public, 5,876 of 9,467 user-run vaults were in loss, together −$123.4 million, against 2,152 in profit, together +$61.9 million (public API, September 2026).
888 algorithms on the Quantopian platform: their backtest Sharpe ratios explained less than 2.5% of how they did live (R² below 0.025).
97% of Brazilians who day-traded index futures for 300 days or more lost money.
Bank for International Settlements researchers estimated that 73% to 81% of crypto app users had probably lost money on their bitcoin.
Of about 300,000 proprietary-trading challenge accounts, about 7% ever received a payout.
An actively managed AI-driven stock fund, AIEQ, returned 4.18% a year over five years, against 13.12% for the S&P 500.
Korea's independent robo-advisor testbed, which tracks real accounts, reported an average of +14.89% for 2025, a year in which the KOSPI rose 75.63%.

The language-model contests

In a widely reported 2025 contest, six large language models each traded $10,000 on a crypto exchange for about two weeks. By the reported final results, two finished up (+22.3% and +4.9%) and four finished down (from −30.8% to −62.7%), and the rankings had changed sharply three days before the end. One two-week run of six players is not evidence of skill in either direction. Our own test of a language model picking trades is here.

And the fraud

"AI trading bot" is now a common scam label. The FBI's 2025 crime report counted 22,364 complaints involving AI, with about $893 million lost. The US Commodity Futures Trading Commission has warned investors directly that AI will not turn trading bots into money machines.

What this means

The automated trading that verifiably makes money relies on speed, being paid to make markets, scale or slow trend following, not on an indicator or a chatbot reading charts.
If a bot seller shows only backtests, short live periods or selected screenshots, treat the claim as unproven. How long proof takes.
Our own six weeks of tests point the same way: the full story.

Sources

Hyperliquid vault figures were read from the exchange's public API on 25 September 2026. Fund and contest results are as reported by the sources above; we did not audit them.