How Vanguard Builds AI It Can Actually Trust With Your Money
The $9 trillion fund giant is wiring AI into its core operations. The woman leading that effort says governance and data infrastructure come before any flashy feature.
Key points
- Vanguard, which manages roughly $9 trillion in assets, is embedding generative AI across its enterprise under a team called the Center for AI and Insights.
- Jing Wang, who leads that team, says five things must work together before AI scales: data, engineering, platforms, operating models and governance.
- Agentic AI systems, software that can complete multi-step tasks on its own without a human clicking through each step, are part of Vanguard's roadmap.
- Heavy financial regulation shapes every deployment decision, making Vanguard's approach a live case study in responsible AI at scale.
- AI2Day has run 28 stories tagged AI governance, 19 of them in the last 90 days; Vanguard's model stands out for putting compliance architecture first, not last.
Most companies bolt governance onto AI after the product is already live. Vanguard's trying to do the opposite, and it's worth paying attention to why.
Jing Wang, who runs Vanguard's Center for AI and Insights, laid out the firm's thinking at MIT Technology Review's EmTech Future conference. The message was direct: you can't scale AI in a regulated environment by moving fast and fixing things later. Data quality, engineering pipelines, the platforms staff actually use, the way teams are organised around AI, and the rules governing all of it have to be designed together, not bolted together afterwards.
What does that mean for ordinary investors?
Vanguard manages money for tens of millions of everyday investors, from workplace retirement accounts to index funds. When a firm that size changes how it processes information, real accounts feel it.
The clearest near-term change is in client experience. AI systems can answer routine questions, flag unusual account activity and personalise guidance at a scale no human team could match. Wang's group is also building toward agentic AI: software agents that handle multi-step tasks (checking a portfolio, spotting a rebalancing opportunity, drafting a summary) without a person managing each click. Done well, that speeds things up for clients. Done carelessly, it introduces errors at scale, and those errors touch real savings.
That risk is exactly why Wang insists governance has to be built into the foundation, not added as a filter at the end.
What makes Vanguard's approach different?
Three things stand out. The firm operates under serious regulatory scrutiny: securities laws, fiduciary rules, data-privacy requirements. Any AI output touching an investment decision carries legal weight, and that pressure forces a discipline most consumer apps simply never face.
Wang's team also sits at the centre of the business rather than off to the side in a skunkworks lab, working across financial services, workplace solutions and corporate services. AI built in isolation rarely survives contact with the rest of an organisation.
Behavioral science sits alongside machine learning, the branch of AI where systems learn patterns from large amounts of data, inside Wang's remit too. Understanding how investors actually make decisions, often under stress and driven by emotion, shapes how tools are designed to help rather than mislead.
Vanguard isn't the only institution thinking this way. Our 5 October story "Enterprise AI Stopped Giving Advice. Now It's Making the Calls." covers exactly the kind of autonomous decision-making Wang is trying to govern carefully, and the gap between those two postures is where most of the real risk lives.
Common questions
Is my Vanguard account already using AI?
Almost certainly in small ways, such as fraud detection and automated customer service routing, though Wang's broader agentic and generative AI programme is still in the build-out phase.
Should I worry about AI making decisions about my investments?
Vanguard's regulatory environment requires human oversight on decisions that affect client accounts, so fully automated investment choices aren't the near-term direction. The more immediate uses are efficiency tools and personalised information, not autonomous portfolio management.
