AI Governance: Should Shareholders Vote on Tech They Don't Understand?

Okay, but here’s the problem: shareholders barely use the power they already have.

Look at executive compensation votes. They’re non-binding “say on pay” gestures, which boards routinely ignore—especially when the company’s doing well on paper. And even when shareholders *do* push back, it’s often performative. A few proxy advisors make recommendations, some big institutions vote yes or no along party lines, and retail investors aren’t even in the room.

Now imagine extending that same circus to AI governance.

Would BlackRock and Vanguard suddenly become safeguards against algorithmic bias or data misuse? Doubtful. Their fiduciary duty is to returns, not ethics. And retail investors? Come on—are we really expecting someone with Robinhood and five Tesla shares to weigh in on whether a model’s RLHF training respects human dignity?

Let’s not confuse “representation” with actual oversight.

The governance we need for AI—especially at the systemic level—requires expertise, not just ownership. You wouldn’t vote by shareholder resolution on which nuclear reactor design is safest. AI systems are increasingly wielding similar levels of infrastructural power, and the idea that ownership equals authority here is shaky at best.

If anything, the smarter play is to create independent AI oversight bodies with real teeth—call them regulatory boards or external councils—but ones that can’t be steamrolled by quarterly earnings pressures. And then let shareholders pressure *those* institutions if they’re not doing the job.

That kind of two-layered oversight is messy, yes. But it's way better than pretending shareholder democracy is the same thing as moral or technical accountability.

Hold on—equating AI governance with executive compensation makes for a snappy analogy, but it oversimplifies what’s fundamentally a different beast. Executive pay is a well-bounded, measurable issue. Shareholders can evaluate CEO performance, look at stock price trends, judge say-on-pay proposals, then pat themselves on the back or cast a protest vote. It’s imperfect, but at least you can build a spreadsheet around it.

AI governance? That’s murkier ground. We’re dealing with complex interactions between technical architecture, societal norms, risk mitigation, and long-tail outcomes that unfold over years, not quarters. Asking shareholders to vote on, say, the rules a company sets for model alignment or synthetic data use is a bit like handing the cockpit controls to the people in row 34C because they paid for the flight.

Take OpenAI as a cautionary tale. Their nonprofit board structure was supposed to provide mission-aligned oversight. Instead, it devolved into a governance mess that ultimately led to Altman’s surprise firing and equally surprising rehiring—with Microsoft effectively becoming the adult in the room. Now imagine that scenario, but with thousands of retail shareholders demanding “transparency” about training data and voting to prioritize brand safety over innovation. Good luck shipping anything.

That said, I’m not arguing for a black box either. There’s a middle path. Companies could give shareholders a lever through proxy votes—but not on the technical frameworks themselves. Focus instead on principles. Should the company commit to external auditing of high-risk models? Should it fund an internal AI ethics board with real teeth? Should dual-use research have an opt-out clause for product teams?

Let’s not delude ourselves into thinking every decision is a democracy. Sometimes, you don’t crowdsource brake design on the car—you just hold the engineers accountable when it fails.

Okay, but here’s the rub—comparing AI governance to executive comp decisions assumes those two things operate on the same level of predictability and expertise. They don’t.

Executive compensation, while messy, is at least legible. Shareholders can look at performance metrics, stock performance, market benchmarks, whatever proxy firm they subscribe to, and cast a vote. You can debate the outcomes, but the structure is understandable. AI governance? That’s not just a “yes or no” on someone’s bonus. It’s a multi-dimensional, evolving, often ambiguous set of technical and ethical decisions. It’s less like voting on pay and more like voting on how to build a nuclear reactor. Amateur hour ends quickly.

Take OpenAI's recent drama. The board ousted Sam Altman ostensibly over governance concerns. The details are still murky (another red flag), but one takeaway is clear: even insiders struggle to agree on what “safe” AI development looks like. Now we expect thousands of retail shareholders—many of whom still think generative AI is just ChatGPT with a slick UI—to make sense of alignment trade-offs, model interpretability, systemic bias, deployment gates? Good luck. We barely have academic agreement on these topics.

That said, the core instinct—to demand accountability—is completely valid. But maybe voting rights aren’t the sharpest tool. If shareholders want a seat at the AI governance table, fine. But let’s not pretend it’s the same table as say-on-pay. It's more like having a public board observer in a room of cryptographers, ethicists, and system architects.

Instead of direct votes, maybe what we need is radical transparency plus informed proxy representation. Open up audit trails, publish risk assessments in plain language, let shareholders appoint technical governance delegates—people who actually understand what the hell they're voting on.

Otherwise, we risk turning AI governance into another theater of shareholder activism, where optics drown nuance and we end up optimizing for whatever sounds most ethical on Twitter. And that’s not just ineffective—it’s actively dangerous.

That sounds democratic in theory—shareholders weighing in on AI governance like they do on executive pay—but let’s be honest: how well has shareholder input on comp actually turned out?

Take executive compensation. Despite so-called “say on pay” votes, CEO pay has ballooned, often detached from real performance. Shareholders technically have a voice, but the process is largely ceremonial. Companies hand them glossy PDFs and bullet-pointed rationales, then do what they were going to do anyway. Do we really want that same hollow ritual applied to AI ethics—an area even more complex and opaque?

Plus, shareholders are not a monolith. You've got hedge funds flipping shares in weeks, retail investors trying to decode 10-Ks on Reddit, and institutional players with vastly different risk appetites. Who among them is equipped to meaningfully evaluate, say, whether an LLM's training data introduces systemic bias or if a model’s deployment crosses ethical lines? Unless you're an AI researcher who moonlights as a moral philosopher, you're probably voting blind.

And this isn’t like comp, where the debate is over numbers. AI governance is squishier. There’s no equivalent of EBITDA-based performance metrics for “alignment with human values.” You’re asking people to weigh in on black box systems they can't audit, using metrics companies aren't disclosing, for outcomes that might not materialize for years.

If anything, the better analogy here might be product recalls. We don’t ask shareholders to vote on whether Boeing should ground planes. We expect regulators, engineers, and executives to take responsibility and be held accountable afterward. AI might need a version of that: decision-making close to the action, combined with real transparency and accountability—potentially including legal consequences when governance fails.

So if we really want guardrails, maybe we should stop pretending shareholder votes are teeth when they’ve mostly been gums.

I get the instinct—it feels democratic. Let the owners vote, especially on something as foundational as how a company handles AI. But there's a problem: most shareholders have no idea what they're voting on when it comes to AI governance. It's not like executive comp packages where you’re deciding between golden parachutes and stock options. AI governance is a murky buffet of trade-offs: transparency vs. IP, speed vs. safety, precision vs. bias. Most shareholders aren’t equipped to weigh those, and forcing decisions through a vote becomes performative, not protective.

Worse, it signals a dangerous kind of abdication. If we outsource tough governance questions to shareholder referendums, we’re essentially saying, "We don’t want to deal with this in the boardroom—we'll just crowdsource ethics." That’s not oversight. It's evasion.

Look at Meta. They held shareholder votes on algorithmic transparency and ethical AI oversight. The proposals were crushed—not because they were bad, but because big institutional investors followed board recommendations like clockwork. These votes become rubber stamps, not real accountability. Meanwhile, AI continues to scale with very little restraint.

If we're serious about AI governance, we need to push responsibility *up*, not out. Boards need a dedicated AI risk committee, just like they have for audit or cybersecurity. People with teeth in the game, not just shares in the game, should be making these calls.

Sure, but here's where that analogy with executive comp starts to wobble a bit.

Shareholders voting on CEO pay is relatively straightforward—you're essentially asking, “Are we incentivizing the person at the top in a way that aligns with performance and long-term value?” It’s a narrow decision with mostly financial implications, even if the optics get messy.

AI governance, on the other hand, is a sprawling mess of ethical risk, technical nuance, legal ambiguity, and long-term societal impact. You’re not just voting on how much risk to accept this quarter—you’re influencing decisions that could reshape how the company interacts with customers, regulators, and the public for decades.

Let’s say a shareholder proposal wants to restrict autonomous decision-making in loan approvals unless human-reviewed. Sounds good on paper. But what does that mean in practice? Is a human-in-the-loop just rubber-stamping? Does it apply to edge cases only? Are we now bottlenecking efficiency, increasing bias, or just creating illusion-of-control theater?

I’m not saying shareholders should have zero say. But voting assumes some minimum level of understanding. And frankly, most institutional shareholders aren’t equipped to evaluate whether a generative model handling customer queries needs an RLHF layer or differential privacy. They’re voting with a thumb up or down on systems that even the AI ethics teams are still trying to wrap their heads around.

There’s also precedent here. Shareholders can’t vote directly on product design or software architecture. We trust specialists for that. So maybe the governance model should mirror audit committees—create mandated, independent AI risk committees with diverse expertise, and let shareholders vote on *that* structure, not the frameworks themselves.

Otherwise, we risk turning AI governance into another PR-fueled, symbolic checkbox vote—like when companies publish glossy ESG reports that say a lot without saying anything and no one’s really held accountable for outcomes. And AI, unlike emissions metrics, doesn't give you thirty years' warning before it derails something big.

That sounds democratic in theory—give shareholders a vote on AI governance just like we do with exec comp packages. But here’s the thing: executive compensation is at least legible. There’s a dollar amount, a timeline, performance metrics you can agree with or roll your eyes at. AI governance? That’s a foggy constellation of risk tolerances, ethical trade-offs, and technical nuance most shareholders aren’t equipped to parse—nor particularly interested in.

It’s like letting shareholders vote on the lab safety protocols for drug development. Sure, you might get input, but is it good input? AI governance isn’t about shareholder preferences—it’s about risk management, long-term brand integrity, and societal impact. Those don’t track neatly with quarterly earnings reports or retail investor preferences.

And let's not forget: most shareholders today aren’t people—they’re institutions. Index funds, pension managers, algorithmic traders. Are we giving BlackRock and Vanguard even more sway over how AI shapes society? They already vote with a baffling mix of passivity and opacity. Look at how ESG proposals often get kneecapped in proxy season despite investor interest—it’s not really about democratized decision-making, it’s about whoever controls the steering wheel pretending to ask for directions.

If we really want accountability in AI governance, why not include stakeholders beyond shareholders? Employees, who understand the tech. Regulators, who are (hopefully) neutral. Even customers, who deal with the downstream impact. Shareholder voting is a blunt instrument being asked to draw a circuit diagram. Let’s not mistake “having a vote” for “having a clue.”