The broken ballot

This is article 1 of 5 in The Broken Ballot, a series on AI, Deliberation, and Responsible Corporate Governance published every Thursday.

A century-old instrument facing a 21st century problem

Picture the annual general meeting of a large listed company. Shareholders arrive, or more commonly cast proxies in advance, on resolutions that may have been circulated in documents running to several hundred pages. Debate is compressed into minutes. The outcome is typically determined before the meeting opens, proxy advisors have spoken, institutional voting algorithms have processed the recommendations, and the formal assembly functions largely as a ritual of legitimation rather than a process of genuine deliberation. Then comes the vote: For. Against. Abstain.

This instrument was not designed for the world in which it now operates. The annual general meeting evolved in the nineteenth century to serve a limited class of owner-managers deciding on dividends and board appointments in a largely domestic economy. Today, the same instrument is applied to decisions involving transnational supply chains, intergenerational climate risk, contested regulatory frameworks, and competing conceptions of corporate purpose. The mismatch is not marginal: it is foundational.

Three structural failures

The inadequacy is not simply about information. Shareholder meetings routinely present vast volumes of disclosure with inadequate time or framework for synthesis. That is a problem, but it is a solvable one: more disclosure, better formatted, earlier distributed.

The deeper failures are structural. First, voting incentives are tilted toward the short term in ways that are built into the ownership model itself: index funds with diversified portfolios and minimal cost-per-vote ratios have weak incentives to invest in the quality of deliberation on any individual resolution.

Second, proxy advisory firms exercise disproportionate influence, creating herding dynamics that crowd out independent fiduciary reasoning: institutions processing thousands of votes annually have strong operational incentives to follow recommendations rather than exercise independent judgment.

Third, and most fundamentally, the binary structure of most votes collapses genuinely complex governance choices into a single gesture. A vote on a climate strategy resolution is not a binary question. It involves probability-weighted scenarios, contested ethical trade-offs, intergenerational welfare considerations, and legal risk, none of which a For/ Against/ Abstain ballot was designed to accommodate.

Why this is now a legal question, not just a governance one

For most of the past century, the quality of deliberation behind a shareholder vote was a governance preference rather than a documented legal standard. That framing is evolving. The trajectory established by recent climate litigation (and by the hardening of stewardship code expectations at the FCA and Investment Association) has begun to transform fiduciary quality from a soft preference into a justiciable standard. Recent climate litigation and stewardship developments suggest increasing scrutiny of the quality of fiduciary decision-making, although the precise legal standard remains unsettled. The question of what that judgment requires (and whether current governance architecture supports it) is no longer academic.

The courts of England and Wales have now made exactly this connection explicit. In ClientEarth v Shell [2023] EWHC 1137 (Ch), Trower J held at paragraphs 65–68 that the management of a company such as Shell requires balancing multiple competing considerations, and that this balancing exercise is fundamentally a matter for directors, not judges — a modern restatement of the business judgment principle that maps directly onto what Sir A. Likierman identifies as the exercise of good judgment.

At paragraphs 82–83, Trower J went further: the proper forum for expressing views about directors’ conduct is the general meeting of shareholders. The court is explicitly relying on shareholder democracy as the corrective mechanism for deficient corporate governance. The argument of this series is that this reliance is currently misplaced — not because shareholder democracy is the wrong principle, but because the institutional architecture through which it operates is incapable of supporting the deliberative quality the court’s expectation requires.

What responsible business requires

Responsible business (the capacity of companies to create long-term value while accounting for social, environmental, and institutional consequences) is not a soft add-on to governance. It is the frame within which governance must now operate. And it requires a different kind of institutional capacity: not just the ability to process information, but the ability to deliberate. To weigh competing interests. To hold uncertainty without collapsing it prematurely into a binary choice. To be accountable for the quality of reasoning, not just the outcome of the vote.

Current shareholder architecture was not designed to support any of that. Without deliberate redesign, AI will make the existing failures faster, cheaper, and more opaque.

Deliberation matters because complex decisions rarely fail through lack of data. They fail because relevant information remains disconnected, dissent is suppressed, assumptions go unchallenged, and stakeholders are excluded from consideration. Deliberation is the institutional process through which information becomes judgment. The purpose of the framework proposed in this series is therefore not to generate more data but to improve the conversion of information into accountable collective choice.

"The binary ballot has had its century. The question now is what replaces it.”

The second article asks a more specific question: what would a governance process designed for sound collective judgment actually look like? Drawing on a 2026 review of 28 Normative Ethical Decision-Making Models, the second article sets out a six-stage AI governance architecture that embeds mandatory deliberative review before the vote, and defines the non-negotiable distinction between AI as deliberative instrument and AI as governance authority.