JPMorgan Cuts ISS and Glass Lewis, Deploys Internal AI for $3.9 Trillion Proxy Vote Stack

JPMorgan Asset Management disclosed it has discontinued subscriptions to Institutional Shareholder Services and Glass Lewis, routing proxy votes for $3.9 trillion in client assets through an internally developed AI system instead. The move removes the two firms that have guided institutional voting decisions for three decades, replacing them with a machine-learning tool trained on JPMorgan's historical vote patterns, regulatory filings, and engagement records. The bank confirmed the system went live in November and has processed votes across 1,847 annual meetings in the current proxy season.

The shift follows years of tension between asset managers and proxy advisors over influence, liability, and the one-size recommendations that ISS and Glass Lewis issue to institutional clients. JPMorgan's system ingests company disclosure documents, cross-references them against the bank's stewardship policies, and generates voting recommendations with explainability reports for internal compliance review. The tool does not rely on external data feeds from ISS or Glass Lewis. JPMorgan's stewardship team retains final sign-off, but the AI handles initial triage, flagging outlier proposals and governance red flags without human preprocessing. The bank has not disclosed the model architecture or training corpus, citing competitive concerns.

This matters because proxy advisors occupy a chokepoint in capital allocation. ISS and Glass Lewis recommendations drive an estimated 25-30% of institutional votes on contested items, giving the two firms asymmetric influence over board elections, executive pay, and shareholder proposals. Their business model depends on asset managers outsourcing judgment to a centralized research function. If the largest allocators build internal systems, the advisors lose revenue and relevance. JPMorgan's AUM makes it the fourth-largest institutional investor globally. BlackRock and Vanguard have not announced similar moves, but both have expanded internal stewardship teams in the past eighteen months. The regulatory tailwind is clear: the SEC's 2022 proxy-advisor rule changes reduced liability for asset managers who vote independently, and the Department of Labor's 2023 guidance explicitly encouraged fiduciaries to develop proprietary governance frameworks. JPMorgan is the first to go entirely off-grid.

The second-order effect is vendor consolidation. ISS derived $563 million in revenue from proxy advisory services in 2023, with institutional asset managers representing 68% of that figure. If three more top-ten managers follow JPMorgan, ISS loses baseline recurring revenue and faces margin pressure. Glass Lewis, owned by the Ontario Teachers' Pension Plan, is smaller and more exposed. The firm has not commented. Separately, this creates an IP moat for JPMorgan. The bank can now license its governance AI to smaller asset managers, pension funds, and sovereign wealth funds that lack the engineering resources to build in-house. That turns a cost center into a product line. Worth noting: JPMorgan did not announce this via press release. The disclosure appeared in a stewardship report filed with the UK's Financial Conduct Authority, where the bank is required to explain voting methodology. U.S. clients received no equivalent notice.

Operators should watch for two follow-on events. First, whether BlackRock or Vanguard announce pilot programs for internal proxy AI in their mid-year stewardship updates, expected in June. Second, whether ISS or Glass Lewis attempt to acquire governance-AI startups to defend their position. The proxy-advisory market has seen zero M&A since MSCI bought ISS in 2014. If a deal surfaces in the next six months, it signals existential concern. Separately, allocators should request explainability documentation from JPMorgan if they use the bank as a voting agent. The AI's decision logic is not public, and fiduciary duty requires understanding how votes are cast on client assets.

JPMorgan's stewardship team processed 412,000 individual ballot items in 2024. The AI now handles 91% of initial reviews, with humans adjudicating the remaining 9% flagged as non-standard. The bank has not disclosed error rates or appeals, but internal compliance requires quarterly audits of AI-generated votes against manual benchmarks. The system is trained exclusively on JPMorgan data, meaning it replicates the bank's historical voting patterns unless stewardship policy explicitly changes. That introduces path dependence: the AI optimizes for continuity, not independent judgment. The tool is not yet deployed across JPMorgan's private banking or wealth management divisions, which still rely on ISS for smaller client accounts. Full rollout is scheduled for Q3 2025, pending regulatory sign-off in twelve non-U.S. jurisdictions.