Some of it is. Most of what’s marketed under the name isn’t — either the AI is a moving-average rule with a new label, or the returns exist only in a backtest, or both.
The field is legitimate. The marketing mostly isn’t. Telling them apart is the skill.
What’s real
Quantitative hedge funds have used machine learning in production for two decades. Renaissance, Two Sigma and D.E. Shaw are AI trading firms in every meaningful sense, and their results are audited. Models that predict short-horizon returns, classify market regimes and optimise execution are standard tools, not speculation.
What isn’t
“AI” as a synonym for automated. A grid bot with an AI sticker is a grid bot. Backtested returns presented as track records. Any promise of a return figure. Any product that won’t show live performance, net of costs, dated, with drawdowns.
The questions that separate them
Is the performance live or backtested, and is it labelled? Is it net of fees and slippage? How long is the record, and did it include a bad regime? What’s the worst drawdown, and how long did recovery take? Can you verify any of it independently — on-chain, through a custodian, through an auditor?
A legitimate operation answers all five without hesitation. An illegitimate one answers with a return figure.
What legit AI trading looks like from outside
Modest claims. Published methodology at the level of approach if not code. Live records with losing months in them. Capacity limits, because real edges are finite. And an explanation of why the edge exists that a sceptical reader can follow.