JPMorgan Chase has discontinued all contracts with external proxy advisory firms and built a proprietary AI-driven system to evaluate corporate governance votes across its equity holdings. The shift affects voting decisions on roughly 12,000 public company proxies annually, spanning markets where the bank holds equity positions exceeding $600 billion in aggregate market value. Final voting authority remains with human portfolio managers and governance specialists, but the machine learning layer now handles initial assessment, flagging, and pattern recognition across shareholder proposals.

The bank disclosed the policy change in a governance filing last week, noting that the internal tool became operational in Q4 2024 after 18 months of development. JPMorgan Asset Management previously relied on ISS and Glass Lewis for proxy research and voting recommendations, a standard arrangement among institutional allocators. The new system ingests public filings, earnings transcripts, board composition data, and historical voting patterns to generate preliminary assessments on director elections, executive compensation, and shareholder proposals. The bank declined to specify the model architecture but confirmed it was trained on proprietary data sets and does not use third-party large language models for final recommendations.

The move carries weight beyond operational efficiency. Proxy advisors have faced sustained criticism from corporate boards and activist investors for wielding influence without accountability, issuing recommendations that shape billions in voting decisions with limited transparency into methodology. By building internal capacity, JPMorgan eliminates the advisory layer that sits between its fiduciary duty and vote execution, gaining speed and reducing the risk of misaligned recommendations on contentious governance matters. The bank also avoids the disclosure lag inherent in advisor reports, which can arrive days before a vote deadline and leave little room for nuanced judgment on complex proposals.

The technical stack matters here. JPMorgan's system flags director nominees with overlapping board commitments, identifies compensation structures misaligned with peer groups, and screens environmental and social proposals for materiality to long-term returns. It does not automate voting on shareholder resolutions tied to political or social issues, where the bank maintains a manual review process. The AI layer compresses analysis time on routine matters from 72 hours to under four hours, freeing governance teams to focus on contested elections and activist campaigns. The bank expects to cut proxy-related operating costs by roughly 30% in 2025 while maintaining the same headcount in its stewardship group.

Other large allocators should watch whether JPMorgan's approach pressures the duopoly of ISS and Glass Lewis. If three more top-ten asset managers build comparable systems in the next 18 months, the proxy advisory business loses structural pricing power and faces margin compression. Smaller allocators without the capital to build internal tools will continue leaning on advisors, but JPMorgan's disclosure may accelerate vendor negotiations and push advisors toward modular, API-accessible research products rather than bundled voting recommendations.

The bank's governance team will publish its first AI-assisted voting record in the 2025 proxy season, covering roughly 2,400 annual meetings between March and June. That data set will clarify whether the internal tool diverges materially from advisor recommendations on director elections and say-on-pay votes, the two categories where institutional voting patterns typically align most closely with ISS guidance.