A new Harvard Law School Forum post by Anteris Advisors describes how AI is becoming the first reader of the proxy statement. J.P. Morgan Asset Management dropped ISS and Glass Lewis for U.S. votes in favor of its in-house AI platform. Wells Fargo moved to a proprietary policy engine. ISS is rolling out PolicySteward for automated, policy-driven recommendations, and Glass Lewis is replacing its single-house view with four research perspectives. Even the SEC’s top investment management official, Brian Daly, says agentic AI in proxy voting “isn’t science fiction—it’s a near-term reality.”
The technology is impressive. The question it leaves open is old: whose interests are being voted?
Automating the Incidental Human
In Incidental Humans: Fiduciary Duty Drift and Restoration, I traced how fiduciary duty began in Chancery as protection for vulnerable people whose property was controlled by others, then drifted toward loyalty to abstractions: the corporation, the fund, freely alienable shares. As Vice Chancellor Laster observed in McRitchie v. Zuckerberg, from the directors’ standpoint, “the ultimate human beneficiaries of that value are incidental.”
The same drift runs through the investment chain. Most Americans own stock through index funds, target-date funds, 401(k)s and pension plans. The intermediaries who vote those shares owe them fiduciary duties. Yet, as Jill E. Fisch and Jeff Schwartz document in Corporate Democracy and the Intermediary Voting Dilemma, and Caleb Griffin in Humanizing Corporate Governance, those intermediaries make little effort to learn what their beneficiaries actually want. Decision-makers assume preferences rather than ask.
Nearly everything in the Anteris post is about helping asset managers apply their own policies faster and more consistently. That’s better than outsourcing judgment to two proxy advisors. But if the policy engine runs only on the manager’s views, AI doesn’t restore the human. It automates the abstraction, at scale.
So how the regulator overseeing fund advisers frames the issue matters. Daly’s AI comment came at the end of a speech whose very title tells us where his focus lies.
What the SEC’s Daly Actually Said
Daly’s January speech to the New York City Bar, (Re)Empowering Fiduciaries in Proxy Voting, is mostly about freeing investment advisers from the “votes have value” and “we will vote all proxies” mantras that, in his view, built an oligopoly of proxy advisors. He notes the SEC’s own 2003 and 2019 guidance allows advisers and clients to agree to limited voting or none at all, and suggests index and systematic funds ask whether voting on precatory proposals fits their passive mandates.
That is a warning for anyone who cares about the humans behind the shares. If index funds stop voting, beneficiaries lose their power. They are silenced, and boards gain power by default.
But Daly also points the other way. He praises fund managers, including index fund managers, who let investors “express their voting preferences and direct their proportionate share of voting power,” calling it exactly the type of innovation regulators should not slow down. His one condition: such programs must offer investors “a sufficient number of options,” with clear disclosure of how shares will be voted.
How many options are sufficient? I would argue − as many as there are investors. A menu of four or five policies is better than one, but the fullest answer to Daly’s condition is a personal voting profile for each beneficiary.
Daly also invokes the 2019 Fiduciary Interpretation’s call for a “reasonable inquiry into the client’s objectives,” asking whether routinely following a proxy advisor’s recommendations meets that standard. It is a fair question. A fairer one: can any fund that votes millions of investors’ shares without asking a single one of them about their objectives claim to have made a reasonable inquiry?
Finally, his AI vision. Daly imagines an agent that can review hundreds of proxy statements, “assess them against your expressed values,” and generate principled recommendations almost for free. He means the adviser’s values. Point the same agent at each beneficiary’s expressed values, and the cost objection to asking investors disappears. AI computing power makes that relatively easy now.
Daly spoke amid an executive order targeting proxy advisors and Chairman Atkins’s call to “de-politicize” shareholder meetings. Granular investor choice answers critics on every side. No one’s ideology is imposed. The votes follow the people whose capital is at stake.
Informed Intermediation Is Now Cheap
Daly’s “reasonable inquiry” points toward what Fisch and Schwartz call informed intermediation. They argue fiduciary duty requires fund managers to make reasonable efforts to identify beneficiary preferences before voting on their behalf. Managers should:
- seek beneficiary input,
- incorporate it into stewardship and votes,
- disclose how they did so, and
- answer to regulators for compliance.
The usual objection has been cost. Surveying millions of beneficiaries on thousands of ballots seemed impractical. The tools in the Anteris post remove that excuse. If an AI engine can apply J.P. Morgan’s custom policy across every ballot it votes, it can apply each investor’s preferences too. Funds can then aggregate those preferences pro rata, as Griffin proposes, or treat them as real input while keeping the discretion Fisch and Schwartz want intermediaries to retain. Either way, discretion becomes relational again.
Recent research with a large Dutch pension fund (Investor Democracy, by Rob Bauer et. al.) shows beneficiaries hold coherent, considered views when given balanced information. In field experiments, randomly selected participants supported expanding impact investments even after being told of possible return trade-offs. Griffin’s survey of 1,611 index fund and ETF owners found similarly prosocial views across income levels and generations. The preferences are there. The duty is to ask.
Beyond a Few Values Boxes
The voting-choice programs at BlackRock, Vanguard, and State Street are a partial correction. They document client preferences, and that matters. But they work mostly as menus: pick one of several third-party policies. Glass Lewis’s four perspectives follow the same logic. Four boxes beat one. They are still boxes.
I’ve tested this. Using iconik, I compared my own S&P 500 votes with those of the Vanguard 500 Index fund (see The Proxy Voting Choice Revolution: Quick Thoughts). None of the menus offered by the largest fund managers would produce anything close. None even offers a portfolio-wide option focused on reducing the externalities that universal owners, which is to say most of us, ultimately pay for (see Reclaiming Fiduciary Duty).
People don’t hold values in bundles. An investor might want strong board accountability, support for employee ownership, skepticism of dual-class stock, and a case-by-case view of climate proposals. Forcing that person to choose between “ESG,” “wealth-focused,” “faith-based” or “board-aligned” captures some of what they think and loses the rest. Menus also let managers say they are just executing client choices, which deflects political criticism without truly engaging the people behind the money.
Brav, Li, Lund and Pan rightly warn in The Proxy Voting Choice Revolution that too many choices can overwhelm rationally apathetic investors. Granular choice doesn’t require answering hundreds of questions. Investors can start from a trusted organization’s policy, then customize where they differ: vote with the Sierra Club on the environment, say, and with Norges Bank or the New York City Comptroller on governance. iconik already allows this.
Nor is granular choice a way to win votes. The same paper found that if Vanguard’s voting-choice program were fully expanded, proposals John Chevedden and I filed at Tesla would have lost. I support it anyway. Fiduciary duty runs to the people whose money it is, whether or not they agree with me.
Fiduciary loyalty to people, rather than abstractions, calls for granular choice: let investors answer the questions they care about, skip the ones they don’t, and have their shares voted accordingly, with the right to override.
It Has Already Been Done
This isn’t hypothetical. In 2023, I wrote about ONEFUND’s partnership with iconik, which let INDEX fund investors build personal voting profiles and gave the fund pro rata voting instructions calculated from them. iconik‘s VoteForge lets users build a profile issue by issue and records how each vote was cast and why. Its partners span the spectrum, from As You Sow and Envestnet to the National Center for Public Policy Research’s Proxy Navigator for center-right investors. The point isn’t to push votes in any direction. It’s to make sure votes reflect the people whose capital is at stake.
Auditability Should Run to People
Daly conditioned his openness to AI voting on human review, transparency, auditability and consistency with fiduciary duties. Those are the right tests, but they should measure more than whether a machine applied the manager’s policy correctly. They should measure whether the fund faithfully carried out, or reasonably weighed, its beneficiaries’ expressed preferences.
Daly asks whether clients should be concerned when an adviser’s votes are nearly identical to a proxy advisor’s standard policies. The same scrutiny belongs on funds whose votes never reflect anything their beneficiaries said. A system that records each investor’s instructions and the resulting vote creates the accountability that “best interest” voting has never had.
This gap is not new. Daly traces today’s voting practices to the Labor Department’s 1988 Avon letter, which declared proxy voting a fiduciary act. In 1995, in my first post on CorpGov.net, I asked the Department whether it had ever acted against a plan fiduciary for voting proxies against participants’ interests. Its Office of Enforcement replied that no such enforcement actions had been taken. Three decades later, Fisch and Schwartz report there appears to be no successful claim that an institutional investor failed the best interest standard when voting. The duty was declared. It has never been enforced. Records that tie each vote to beneficiaries’ instructions would finally make it enforceable.
The Anteris post advises companies to re-map the vote because AI-enabled systems, custom policies and pass-through arrangements make historical voting patterns less predictive. Companies should also get used to the idea that the humans behind the funds may soon be heard directly, one granular preference at a time.
What Should Happen Next
- Funds and pension plans should offer personalized voting profiles, not just menus, and disclose how beneficiary preferences shaped their votes.
- Funds considering not voting should ask their investors first. Silence by default makes beneficiaries more incidental, not less.
- Proxy advisors and AI platforms should build tools that collect and apply beneficiary preferences, not only manager preferences.
- Regulators should treat seeking and reflecting beneficiary preferences as part of fiduciary duty under the Advisers Act and ERISA’s “solely in the interest” mandate, as Fisch and Schwartz propose.
- Individual investors should use the tools already available. Rational apathy isn’t fixed. Lower the cost of voting and more people vote.
- Funds should announce their votes in advance, as I urged in Fiduciary Duty to Announce Votes. AI-assisted analysis makes earlier decisions practical, and beneficiaries and other shareholders deserve to know how their agents intend to vote while it still matters. Plus, those disclosed votes can move other votes. See Proxy Voting at Norges: Announcing Votes Triples Impact.
AI is changing who reads the proxy. Fiduciary duty exists to protect people, not abstractions. The real test of AI voting is whether it makes the humans behind the shares central again, or leaves them more incidental than ever.
What This Means for Rule 14a-8
All of this is happening while the SEC proposes to rescind Rule 14a-8, the rule that requires companies to include qualifying shareholder proposals in their proxy materials. The proposed release never mentions AI-enabled voting, voting-choice programs, pass-through voting, or its own Investment Management Director’s praise for investor-directed voting. It treats proxy voting as if it were standing still. It isn’t.
Automated voting only reaches the proxy card. iconik profiles, voting-choice programs, and AI policy engines all vote on the items in the company’s proxy. Without Rule 14a-8, most shareholder proposals would move to the meeting floor, where few proponents can afford their own solicitation. Investors voting through these systems would never see them. Worse, the SEC’s companion amendments to Rule 14a-4(c) would let management cast discretionary votes on those floor proposals unless shareholders opt out. Rescission would shift voting power from beneficial owners to management, just as technology makes it practical for the real beneficial owners to be heard.
The SEC is contradicting itself. Daly encourages funds to let investors direct their votes and to make a “reasonable inquiry into the client’s objectives.” The Commission would, at the same time, remove the shareholder-initiated items that investors would direct votes on.
Low support measures the agency problem. The release leans on the fact that most proposals fail. But those votes are cast mostly by intermediaries who never ask their beneficiaries. When people are asked, as in Griffin’s survey and the Dutch pension fund’s binding vote, they express considered preferences that today’s vote counts don’t capture. As pass-through and AI voting spread, the historical data the release relies on becomes less predictive, a point Anteris makes to companies directly.
The cost picture is outdated. The same AI tools that help investors analyze proxies help companies evaluate and respond to proposals. The release’s cost estimates predate them.
Proposals put investors’ words in the proxy. If machines now read the proxy first, shareholder proposals and supporting statements are among the few places where investors, not management, write what those machines read.
Fiduciary duty exists to protect people. The tools to let people’s voices count are finally here. Rule 14a-8 keeps the questions they care about on the ballot. Comments on the rescission (File No. S7-2026-32) are due November 20, 2026. I am filing a supplemental comment making these points, and I encourage others to do the same.
Disclosure: I am an investor in iconik and use it to vote my own shares. I also hold a small position in ONEFUND’s INDEX.
Collaborate with CorpGov.net on Social Media (or Leave Comments at the bottom)
Linkedin
BlueSky
Facebook
Mastodon
Substack
X (formerly Twitter)

No comments yet.