JPMorgan Just Fired ISS and Glass Lewis, Replacing Them With AI

The AI Proxy Advisor: How Institutional Investors Are Ditching ISS and Glass Lewis

The Duopoly Meets an Algorithm

The proxy-voting duopoly that has directed trillions in shareholder votes for decades just lost its biggest client to an algorithm.

JPMorgan Chase's asset-management division, overseeing more than $7 trillion in client assets, CNBC reported, terminated its contracts with Institutional Shareholder Services and Glass Lewis in January 2026 and replaced them with an in-house AI platform called Proxy IQ. The tool aggregates and analyzes proprietary data from more than 3,000 annual company meetings, producing vote recommendations for portfolio managers on the firm's Spectrum investment platform, which carried over $3 trillion across more than 11,000 portfolios at the end of 2024, OpenDataScience's data shows. A company spokesperson said JPMorgan will complete the transition during the first quarter of 2026, framing the move as the first time a major investment firm has fully eliminated external proxy advisers from its U.S. voting process. The AI-governance market driving this shift is growing at roughly 30 percent annually, per multiple research firms tracking the category through 2035.

Political pressure accelerated the break. In December 2025, President Trump signed an executive order directing the SEC, Department of Labor, and FTC to review proxy-adviser practices, alleging the firms "regularly use their substantial power to advance and prioritize radical politically-motivated agendas" such as DEI and ESG proposals. The order also flagged conflicts of interest and recommendation quality. Jamie Dimon, JPMorgan's CEO, had already called proxy advisers "incompetent" and decried their "undue influence" in his 2023 shareholder letter, stating the firm planned to "generally eliminate third-party proxy advisor voting recommendations from its systems." Texas, Florida, and Mississippi attorneys general have opened parallel probes.

Into that vacuum steps Pennant, a Y Combinator S26 startup founded by Ryan Nowicki Stewart and Tomas Taylor, former institutional proxy voters and technologists with a collective 20-plus years of experience. Pennant's pitch is not just automation but a "world model" that learns the governance logic behind each vote, positioning the product as the scalable alternative every other institutional investor can now buy rather than build.

JPMorgan's $18 billion technology budget let it build Proxy IQ internally, Business Insider found. Pennant is betting the rest of the market (pension funds, asset managers, corporate advisors) will need a vendor that delivers the same capability without a nine-figure engineering org. The duopoly's grip is suddenly contestable.

What a Governance OS Actually Has to Build

A single annual meeting generates a DEF14A proxy statement that can run hundreds of pages, plus 8-K filings, no-action requests, exempt solicitations, and a stream of amendments. Multiply that by thousands of portfolio companies and the volume becomes staggering. The core problem Pennant addresses is not "what should we vote?" — that is a policy question — but "how do we get from raw filings to a defensible vote recommendation without drowning?"

The platform pulls meetings, filings, and the aforementioned data into one workspace. It ingests SEC EDGAR filings, court records from Justia and UniCourt, and data from Bloomberg, FactSet, and Broadridge. It tracks no-action activity, litigation, and proponent behavior over time. Every data point remains tied to its source document: what Pennant calls "source-linked intelligence." That traceability matters when a compliance officer or auditor asks why a vote went a certain way.

The recommendation engine is a starting point, not a verdict. AI handles the grunt work — reading 500-page proxies, comparing precedent, flagging outliers — so analysts can focus on the meetings that matter. The workflow translates a firm's voting guidelines into recommendations with clear rationale and human review. It flags pay-for-performance concerns — "CEO pay jumped 40% and returns were flat" — but leaves the final vote to the team. A governance analyst might flag a director with 14 years of tenure as above a 12-year independence threshold. The head of stewardship then notes that after direct engagement, they remain confident supporting the director. The chief stewardship officer approves the override, with rationale on file. AI surfaces the risk; humans make the call.

For compliance teams, the killer feature is the N-PX export. Mutual funds must file proxy voting records annually on Form N-PX. Producing that filing manually is a nightmare of spreadsheet wrangling. Pennant automates the export and maintains a complete vote record trail: exactly what an auditor wants to see. The platform captures sources, comments, approvals, overrides, and rationale from intake to final decision.

Research, internal notes, and voting rules often sit scattered across filings, vendor systems, and spreadsheets. Pennant sells a single layer for the research and decision process. Its "world model" branding is broader than the product available today; the current platform is a focused set of research, policy, and recordkeeping tools for proxy voting. But the architecture bets that investors, public companies, and advisors can share the same underlying data layer (proposals, proponents, precedents) and merely see different views. That means Pennant is selling to three different buying committees, each with its own procurement cycle. Custom pricing for all three tiers suggests the company is still in the land-and-expand phase, tailoring scope per client.

Traditional proxy advisors like ISS and Glass Lewis provide voting recommendations. Pennant does not. It provides the infrastructure for investors to make their own recommendations, aligned with their own principles. That distinction matters in a market where asset managers face pressure to demonstrate independent judgment. Public companies, law firms, and bankers can also use the platform to track proposals, research precedents, study a shareholder base, and prepare for annual meetings, though Pennant must maintain clear controls between investors making voting decisions and advisors or companies attempting to influence those decisions.

The Market Behind the Disruption

The AI governance category has graduated from pilot budgets to line-item spend. Multiple research firms converge on roughly 30 percent annual growth through 2035, though their starting points differ.

The variance reflects different boundary definitions: some count only dedicated governance platforms, others include adjacent risk and compliance tooling. North America generated $392.9 million in 2025, nearly half of global revenue, according to OpenDataScience. The banking and financial-services vertical alone accounted for more than a quarter of spend. Large enterprises command nearly three-quarters of the market today, but small and mid-size firms are growing at 44 percent annually, the fastest segment. Services are outpacing software at 41 percent, a signal that implementation and ongoing oversight are where the marginal dollar flows.

Three forces drive the curve. First, regulation has moved from voluntary frameworks to binding law. The EU AI Act entered force in August 2024 with phased obligations for high-risk systems hitting by August 2026. The OECD AI Principles, updated in 2024, now inform policy across more than 70 jurisdictions. NIST's AI Risk Management Framework and its Generative AI Profile added lifecycle-oriented governance requirements, particularly for systems whose risk profile shifts post-deployment. The UK Financial Conduct Authority published final guidance in April 2025 requiring algorithmic transparency in consumer-facing AI by Q1 2026.

Second, enterprise AI deployment has outrun the manual controls that once sufficed. Stanford's AI Index recorded global corporate investment in AI ethics and compliance tools surpassing $1.4 billion in 2024, Stanford's AI Index found, double the figure from two years prior. Legacy processes (spreadsheets tracking model versions, ad-hoc fairness checks, email-based approval chains) are being replaced by integrated platforms that automate bias detection, drift monitoring, and audit logging across the full ML lifecycle. Cloud-native governance suites now embed directly into MLOps pipelines, displacing fragmented point solutions with unified dashboards.

Third, the proxy-voting duopoly itself is a market accelerant. ISS and Glass Lewis together hold over 90 percent of the proxy advisory market, according to the White House. That concentration has drawn sustained political scrutiny: lawmakers and regulators argue the duopoly wields undue influence over corporate governance outcomes. The administration has challenged the power of proxy advisors directly, and Skadden notes this pressure makes it harder to muster support for critical shareholder votes while outcomes become less predictable. At the same time, institutional investors are pulling voting decisions in-house, as Harvard Law's Corporate Governance forum documents the shift toward customized voting policies developed with AI, away from benchmark policies supplied by the advisors.

The incumbents' moat is narrowing from both sides. Regulators are demanding transparency around how voting recommendations are constructed, especially where DEI and ESG factors weigh in. Investors are discovering they can ingest SEC filings, meeting materials, and engagement records directly, map them to their own policies, and produce audit-ready rationales without an intermediary's black box. The addressable spend is broadening because model inventories alone don't close governance gaps: enterprises now require a connected control layer spanning use-case intake, risk classification, documentation, testing, approval workflows, production monitoring, and incident escalation. Financial institutions and healthcare providers are leading the pull for unified platforms that combine bias detection, feature importance analysis, and decision explainability in one stack.

Headwinds remain. High implementation complexity and total cost of ownership shave an estimated 4.2 percentage points off the CAGR for mid-market adopters. A critical shortage of certified AI governance professionals cuts another 3 points. Regulatory fragmentation across jurisdictions drags roughly 12 percent; talent gaps in AI ethics and compliance another 10 percent; the absence of standardized audit benchmarks a further 8 percent. But the vector is clear: the proxy-voting workflow is being re-platformed, and the capital follows the workflow.

Reactions: Regulators, Advisors, and the Fiduciary Question

The regulatory scaffolding that propped up the proxy-advisor duopoly is cracking. In August 2026 the U.S. Department of Justice's Antitrust Division withdrew the 1987 business review letter that had shielded Institutional Shareholder Services from antitrust enforcement, citing concerns that ISS and Glass Lewis wield outsized power to shape corporate governance policies. A December 2025 executive order had already directed the SEC to review whether proxy advisory firms should register as investment advisers and whether investment advisers following proxy recommendations on non-pecuniary factors breach fiduciary duty. The message from Washington is clear: the market share these two firms command is no longer a settled fact of market structure — it is a target.

SEC Division of Investment Management Director Brian Daly addressed the shift directly in January 2026. He affirmed that fiduciary duty includes monitoring corporate actions and voting proxies, but also acknowledged that not voting "makes sense in many situations," urging flexibility between advisers and clients. Daly pointed to large language models and agentic AI as "compelling opportunities" for handling proxy voting's scale and complexity, then added the condition that will define the next phase: AI must be "done right," with proper training and oversight to align with fiduciary duties. He closed by urging investment advisers dissatisfied with the current landscape to "re-evaluate and reassess" their proxy voting strategies (regulator-speak for permission to build in-house).

Glass Lewis read the room. In April 2026 it published a fiduciary-test guide framing five questions every asset manager should ask: what governs the underlying data, what the human actually does, how "investment-grade" is defined, what the AI's designed scope is, and what the accountability structure for exceptions looks like. The firm also announced it will eliminate its standard benchmark voting recommendations in 2027, replacing them with recommendations built on each client's specific investment philosophy. The pivot is both a product move and a defensive signal: Glass Lewis is betting that the future belongs to customized policy engines, not one-size-fits-all templates — and it intends to sell the engine.

Law firms are tracking the same vectors. Skadden's September 2026 activism update warned that continued regulatory scrutiny of proxy advisors and efforts to curb voting blocs at large institutions may make shareholder votes harder to predict. The "Big Three" asset managers have split their stewardship divisions, raising the prospect of split votes within a single institution, while expanded pass-through voting programs let underlying investors direct decisions instead of central stewardship teams. JPMorgan and Wells Fargo have already cut ties with proxy advisory firms, opting for AI-assisted internal voting. The infrastructure of influence is fragmenting in real time.

The fiduciary question now centers on auditability. Harvard Law School's March 2026 research on proxy-research workflows found that general-purpose models score roughly 84 percent accuracy on proxy statements, insufficient for a function where errors cascade into votes, engagement positions, and client reports. The validated benchmark for production systems is 99.2 percent, achieved only when every output traces back to a triggered rule and the underlying source document. Stewardship teams can now generate fully sourced "For," "Against," or "Refer" outputs within minutes of a meeting announcement, but only if the platform guarantees that chain of custody. ProxyBeacon's approach (plain-language policy definition, market- or sector-level application, backtesting against historical data) shows how the cost of iteration drops when the tooling is built for self-serve governance rather than advisory dependence.

Transparency is the unresolved tension. Glass Lewis insists human oversight is an expected baseline; the SEC says AI must be "done right" but has not defined the standard. No regulator has mandated explainability thresholds for voting engines, and no court has tested whether an algorithmic rationale satisfies the duty of care. Until that framework hardens, every in-house AI vote carries a latent legal risk: a challenger could argue the model's logic was opaque, its training data biased, or its exception-handling absent. The firms moving fastest — JPMorgan with Proxy IQ, Pennant with its governance OS — are effectively writing the standard by shipping it.

Whether AI voting scales or stalls depends on which side of that risk the next enforcement action lands.

Who Gets Hired Next

Pennant, a two-person Y Combinator-backed startup, is building out its engineering team to design, build, and ship the core AI features that ingest SEC filings, map proposals to custom voting policies, and output audit-ready rationales. The company's Governance Arena, a public battle-style comparison tool that lets users pit voting scenarios against each other, generates labeled training data for the world model Stewart and Taylor describe.

The market data bears out the breadth of that demand. LinkedIn shows 67,000 open AI Governance roles in the United States as of the latest count, with a parallel 53,000 listings under "Artificial Intelligence Governance" and 9,000 more on Indeed. Proxy-voting specific postings are thinner (74 on LinkedIn) but the job descriptions reveal the hybrid profile: 12-plus years executing proxy voting and engagement research inside asset management, direct shareholder engagement with public companies, and policy development that can survive a SEC exam. Stewardship teams at firms like Parnassus describe their toolkit as three levers — company engagements, proxy voting, and advocacy.

Compensation benchmarks from frontier-tech employers confirm what that hybrid profile commands. Databricks, which added 32 roles in the past week, bands its salaried positions at $140k–$318k with a $250k median. Anthropic, adding 50 roles in the same window, runs $205k–$563k with a $395k median. Pennant's founding-engineer slot will likely price above the Databricks median and toward the Anthropic floor, given the domain scarcity.

The hiring wave extends beyond the startup. The skill set is consistent: fluency in SEC filing schemas (DEF 14A, 8-K, 13D), experience fine-tuning models on sparse, high-stakes text, and the ability to write the policy-to-code translation layer that auditors will inspect. Law firms like Skadden and the Harvard Law School Forum on Corporate Governance have started publishing analysis on AI in proxy voting workflows, drafting the compliance framework the engineers must satisfy.

For engineers who want their model weights to move markets, that is the differentiator.

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.