Articles on AI in proxy voting and stewardship, from the founders and the advisory board. Each one can be read here or downloaded as a PDF.
Nobody publishes a quality KPI for AI-based proxy research, and a backtest against actual votes is a proxy, not an answer key. We propose an evaluation framework scored against a human-built golden dataset, share first results and invite other providers to join.
Committee memberships are still hard for AI agents to read from proxy statements: they sit in tables, biographies, footnotes and pictures. Three changes issuers can make on their own, without a new rule.
Two reports on AI in proxy voting landed in April 2026, both with merit, both reaching conclusions we challenge. Volatility and press-release reliance are properties of naive prompting, not of AI applied inside a policy-first architecture.
By April, stewardship teams are in the tunnel. The constraint is not effort but timing: external research arrives weeks after the filing, so the most consequential decisions get made under the greatest time pressure.
Proxy voting is one of the most structured ways shareholders exercise governance rights, yet applying a policy consistently across thousands of ballot items remains costly. The design choices that make AI-assisted voting traceable, consistent and auditable.
