The best AI funds keep the firmest human veto. What they automated, and what they refused to, shows how AI actually works.AI 用得最好的那几家基金，把人工否决权攥得最紧。 它们把什么交给了 AI、又拒绝交出什么，正好说明 AI 到底是怎么起作用的。

Business Insider reported on Oct 28 that Citadel, Bridgewater and Man Group are pushing AI deeper into their work while leaving a person at the controls. Citadel's assistant reads filings and broker research for nearly all its equity investors, but the trade stays with the human. Man Group runs its language-model workflow under oversight and notes the models are strongest on public information, which is the part already priced in. Bridgewater will let a system run a strategy, then make a portfolio manager sign off.

Automating a process does not retire the judgment. It relocates it: to whoever designs the guardrails, and to whoever decides when to override the output. The step AI took is synthesis. The step it left you is deciding what to disagree with.

These firms also see the failure mode that matters. It is not one rogue bot. It is everyone renting the same good model, trained on the same public signals, reaching for the same trade, so a normal selloff clears in lockstep. That already happened this year.

If the people with the deepest models won't hand over the decision, think twice before handing yours to a chatbot. The edge was never the model. It is knowing which step still needs you.

This breaks if these funds drop the human sign-off and let the model execute on its own.

Research, not investment advice.