When my AI agent just holds shares, who actually earns the securities lending fee?

This week was a good week to be holding rather than trading. The Nasdaq Composite fell 4.6% over the rolling week through June 27, SpaceX cooled from an IPO high of $226 to about $153, and Broadcom gave back most of a 40% gain to sit around 9% up for the year. An AI portfolio agent running a buy-and-hold mandate spent the week doing almost nothing. The positions sat there. Which raises a question I have started asking about my own statements: while those shares sit there doing nothing, is anyone earning money on them, and is it me?

For a growing number of us the answer is yes, someone is, and the someone is usually the broker. The mechanism is fully paid securities lending, and it is worth understanding exactly, because it is the rare income stream that shows up on shares no one is trading at all.

How a borrow fee on your idle shares gets split before it reaches you

Here is what securities lending is, stripped to the plumbing. When an investor fully owns shares (no margin loan against them), the broker can borrow those shares from the account and lend them on to someone who wants to sell short. The short seller posts collateral, usually cash, and pays a daily borrow fee for the privilege. That fee is the price of locating hard-to-find stock, and on heavily shorted names it can run from a fraction of a percent a year to double digits.

The broker collects that gross fee, reinvests the cash collateral for a little more, and then pays the client a contractually fixed fraction of the proceeds. The fraction is where it gets interesting. Charles Schwab splits the income 50/50 with the client. Fidelity and Robinhood land in the same neighborhood, sharing at least half. But M1 states plainly that you “automatically earn 10% of the revenue generated from lending out your shares,” and Public sits at a similar 10%. So the same loaned share, at the same borrow rate, pays one investor five times what it pays another, purely on the split.

Why the gross rate stays invisible until 2027

Now the part that actually decides whether this is a fair deal. We see the amount credited to the account each month. We do not see the gross borrow fee the broker collected, so we cannot compute our own split. There is no benchmark on the statement, no rate disclosure, nothing to check the 10% or the 50% against. The reason is structural: per-loan rate data only becomes publicly reported under the SEC’s securities lending transparency rule, adopted in October 2023, which pushes loan-level reporting to a new engine run by FINRA (the broker-dealer self-regulator, whose data feed is the only place these rates will ever surface for the public) with a compliance date of September 2026 and public dissemination not arriving until March 2027. Until then, the denominator is dark.

That darkness is not hypothetical. The regulator already fined one clearing firm for exactly this gap.

The $3.2 million settlement was about misrepresenting what investors would receive for their loans, across roughly 5 million enrolled accounts. The lesson is not that lending is a scam. It is that when one side knows the gross fee and the other side does not, the disclosure of the split is the entire game.

This is the agentic twist on a very old mechanism. Securities lending generated more than $15 billion for lenders in 2025 by EquiLend’s count, and the retail brokerage segment is its fastest-growing corner. None of that revenue is captured by any fill report or execution-quality statistic we might pull, because those only ever describe trades. Lending income is the yield on not trading, and an agent that trades less makes more of it.

Lending income is the yield on not trading, and an AI agent that holds rather than churns makes more of it, on shares no fill report will ever mention.

What this changes about evaluating a holding agent

The decision lens here is tool fit, and it is narrow but real. When comparing brokers that an AI agent might run inside, the lending split belongs on the same line as commissions and the cash-sweep rate. A 50% program on a portfolio with a few hard-to-borrow names can quietly add more than the fee savings we usually chase, while a 10% program hands most of it to the house. The question to put to a broker is not “do you offer lending,” it is “what fraction of the gross fee reaches me, and how would I ever verify it before 2027.” If they cannot answer the second half cleanly, that itself is the answer.

I keep coming back to one quiet asymmetry. The agent is built to be patient, and patience is supposed to be the cheap virtue in investing. Here it has a price, paid to whoever owns the part of the trade we cannot see. The interesting question is not whether to lend the shares. It is whether the patient agent we hired to do less was ever told that doing less was also a way of earning, for someone.