AI Proxy Advisors Are Already Voting Against Management: What Your Board Needs to Test Now
Should you start testing your proxy score now?
The model scoring your next proxy fight hasn't been validated, tested, or governed by anyone yet.
The Situation
J.P. Morgan discontinued its ISS and Glass Lewis subscriptions in January 2026, publicly, and replaced them with a proprietary AI voting engine named Proxy IQ. Kekst CNC then tested four major large language models as stand-in proxy advisors against contested annual meetings held between 2023 and 2025. Every model recommended the activist's slate or a split card in 59-77% of the contests tested (77% was Gemini), against 55-56% management support historically recorded by ISS and Glass Lewis over the same period.
The Exposure
The Harvard Law School published a study by Kekst CNC which found AI backs activists most often on underperforming share price or total shareholder return and inadequate governance protections. Company-issued press releases supplied nearly one-third of all citations feeding the recommendations, which means the AI evaluating your proxy fight is deciding based largely on what you and the activist each choose to publish. Activist investors are already using these same models to generate campaign content and shape their public arguments directly from voting disclosures, so the case against you may already be tested against an AI audience before your board ever sees it. J.P. Morgan didn't quietly swap its subscriptions, it did it publicly, and a flagship institution normalizing the practice is exactly the “permission” that other funds watch for and follow. The same analysis found that the model doing that scoring isn't stable either, with re-runs of the identical query changing the vote in 38% of the contests tested.
The Judgment Call
Most boards have no idea an AI model may already be forming a view on their next proxy fight, and even fewer know what that view is based on. This isn't a slow-moving trend just being tested, it's already live inside one of the largest institutional holders in the market. The board-level gap is in basic awareness, and the technology has moved so fast that most governance programs haven’t yet caught up. Kekst CNC's data shows the models doing the scoring aren’t stable, let alone validated, since re-runs of the identical query changed the vote 38% of the time. Your next move isn't guessing which arguments an AI evaluator finds persuasive, it's asking the same models the questions activists are already asking about your company, logging what comes back, and controlling that record the way you'd control any other AI system output that could matter to a fiduciary duty inquiry.
Risk: Testing your company against one or two AI models and documenting the results creates a paper trail that could just as easily surface a bad score as a good one, which puts additional pressure on remediation.
Benefit: Seeing the specific factors driving activist and shareholder voting decisions gives the board concrete input for its broader AI governance mandate, and a documented basis for meeting fiduciary duty standards.
This Week’s Action
What to do: Confirm which AI model, or models, your company is using under an enterprise agreement that guarantees the provider isn't retaining your prompts or outputs for training to ensure confidentiality. Then compile your last 12 months of SEC filings, press releases, and any activist activity or focus areas connected to your company, and run that material through the model several times, asking it to assess your proxy vulnerability the way an activist-aligned evaluator would.
Who to involve: General Counsel and Head of Investor Relations.
What outcome to achieve: A documented set of responses across multiple runs, showing the range of the model's reasoning rather than a single output, reviewed and filed by counsel.
Time required: 60 minutes to compile the source material and run the queries across several iterations, 45 minutes with counsel to review the range of results.
Artifact
Use this to interpret whatever comes back from the exercise above, scored against what Kekst CNC found actually moves AI's recommendations. Two or more "No" answers is where an AI-driven activist argument is likely to go first.
Step 1: TSR and Share Price Narrative. Have you proactively addressed any share price or TSR underperformance in your own press releases, rather than leaving that story for an activist to write?
→ Yes: Proceed to Step 2.
→ No: Stop. Underperforming share price or TSR was the most cited reason AI backed an activist's case for change.
Step 2: Governance Protections. Do your recent disclosures affirmatively describe your shareholder protections and governance policies, rather than leaving that ground uncontested?
→ Yes: Proceed to Step 3.
→ No: Stop. Inadequate governance protections was the second most common reason AI recommended supporting an activist.
Step 3: Proxy Advisor Alignment. If ISS or Glass Lewis has ever sided with management on a comparable matter, has that recommendation been publicized somewhere AI can cite it?
→ Yes: Proceed to Step 4.
→ No: Stop. ISS and Glass Lewis recommendations were a contributing factor in AI's reasoning.
Step 4: Owned Content Volume. In the last 90 days, has your team issued press releases addressing strategy and operational progress, rather than relying on fight decks or media coverage to carry that message?
→ Yes: Cleared. Repeat this test whenever a new major AI model releases or a contest becomes likely.
→ No: Stop. Press releases supplied nearly a third of all citations feeding AI's recommendations, more than top-tier media and fight decks combined.
If you don't have the internal bandwidth to structure this analysis, write the queries, or interpret a spread of results that don't agree with each other, that's exactly the kind of project an external advisor with AI governance expertise is built to scope.
When the stakes exceed your internal capacity:
AI Exposure Diagnostic: A 2-hour strategic evaluation for risk, compliance, and legal leaders to identify your highest-priority governance gaps and deliver a 90-day remediation roadmap.
12-Week Governance Sprint: Translate regulatory requirements into audit-ready policies, control frameworks, and accountability structures.
Ongoing Advisory Retainer: Embedded judgment for policy updates, vendor assessments, and board prep as regulations and technology evolve.
Reply with "Diagnostic" or “Sprint” to schedule a conversation for next month.
Chris Cook writes Judgment Call weekly for compliance and risk officers navigating AI governance.
Former IBM Vice President and Deputy Chief Auditor. Published in the AI Journal, speaker at Yale.
Chris Cook
Managing Partner & Founder
Blackbox Zero
Forwarded this by a colleague?Subscribe to Judgment Call
