INDUSTRY PERSPECTIVE

Front Door or Side Door?

An FDX View on How AI Tools Reach Financial Data

NOVEMBER 2026

A growing number of AI-powered tools are seeking access to consumer-permissioned financial data, as more consumers use AI to help manage their money. The question is: Which door will these tools use to access the data?

Many AI-powered tools today fetch this data via secure APIs purpose-built for third-party access. We’ll call this the front door. Others are configured to use channels designed for humans (e.g., bank websites), logging in with the consumer’s credentials, in a practice often described as ‘screen-scraping’ (the ‘side door’).

Purpose-built APIs—where available—offer clear benefits over screen-scraping. They help AI-powered tool providers and agent operators deliver innovative experiences; they help financial institutions protect customers; and they give consumers more transparency control over their financial data.

What the past decade has taught us

While the rise of AI-powered tools raises several genuinely novel and important questions around data access and API design (more on that below), much of what’s needed to deliver safe connectivity already exists.

The financial services industry has made major strides in making APIs the preferred channel for user-permissioned data sharing in recent years. As a result, most consumers in America today can share their financial data via standardized APIs. This migration off screen-scraping has helped fuel a robust ecosystem of thousands of apps that use permissioned data to serve consumer needs.

This progress was made possible in part because of industry coordination anchored around three core tenets. When the Financial Data Exchange (FDX) was founded in 2018—as a broad coalition of banks, technology companies, data aggregators, and others—the vision to promote more secure financial data access was rooted in the beliefs that:

  1. Methods of access should be designed around the principles of Control, Access, Transparency, Traceability, and Security.
  2. Industry standards help companies enable third-party data access because interoperability matters for building cost-efficient and secure, trusted connections at scale across a diverse, multi-party ecosystem.
  3. Consensus-building among diverse industry players supports the creation of standards that ultimately work well over the long run.

These core tenets are equally relevant today as the industry works to adapt secure access methods for AI-powered tools.

What makes for a good front door? Foundational features not to lose

How can the industry ensure it is moving forward, not backward, when developing or deciding on the front-door methods by which AI-powered tools access financial data?

A good starting point is to look at several features that already exist in the APIs used today by thousands of financial institutions. While access methods and standards will evolve to meet new requirements unique to AI tools, the features below represent a strong baseline to preserve and build on. By and large, these are features that “screen scraping” access methods fail to deliver.

Security & traceability

  • No credential sharing: Consumers don’t need to share their (e.g., bank) login credentials with a third-party tool to enable data access.
  • Onboarding & traceability: All parties in the data-sharing chain know each other and can see which data is accessed, by whom, and when. This supports customer protection, transparency in consent, and remediation if something goes wrong.
  • Infrastructure security: Even if a bad actor compromises part of the connection, layered security safeguards keep them from reaching the consumer’s data.

Control & transparency

  • Specific consent: The consumer can give the third-party tool access to only those accounts and data types the consumer chooses and that are needed to enable the service.
  • Ongoing visibility & control: Consumers can easily see (e.g., via a dashboard) which data the third-party tool is able to access, and they can revoke access at any time. (In many cases, the consumer may look to their financial institution for this feature.)
  • Governed consent that stays in sync: When a consumer gives or removes consent, every party in the chain can be updated quickly and reflect that change accurately in their service channels. Additionally, consent is defined, recorded, and enforced through a common framework, rather than left to each party to interpret on its own.

Reliable access

  • Data accuracy: The third-party tool can accurately interpret the data because it is shared in a structured, deterministic format with a defined meaning. This helps improve decisioning, reduce customer complaints, and avoid broken experiences for consumers.
  • Planned capacity: Third-party access requests come through a channel built to expect them, so all parties can plan the infrastructure capacity needed to keep access reliable for all customers.
  • Stability: The channel changes under governance and change management, so existing connections keep working as it evolves.
  • Purpose-built functionality: The channel supports the specific user journeys that the financial institution and the involved third party(s) have agreed to deliver securely & reliably. This bolsters confidence for the consumer, who may look to any one of these parties for help or answers.

What needs to change when AI tools and agents are the ones accessing the data?

Financial data-sharing APIs and standards may need to evolve to support the variety of new ways in which permissioned financial data reaches AI-powered tools.

Considerable work is underway at FDX to support that evolution. This includes work on agent identity, traceability, and onboarding; data semantics and accuracy when AI-powered tools interpret data and documentation; how to safely share data through agent tooling interfaces (where protocol layers such as MCP sit between an agent and an API); and consent journeys. FDX is considering these new features for future versions of our API specification.

Ways to get involved in FDX’s work on AI data access include joining the FDX AI Task Force to influence updates to FDX standards; attending the FDX Global Summit in March alongside industry experts; and subscribing for updates on FDX’s AI work. Visit FDX’s Agentic AI hub to learn more.

Access by AI tools to sensitive financial data may bring different risks, customer expectations, or regulatory impacts than when AI tools access other types of data. As such, our focus at FDX remains on creating and evolving a technical standard that is tailored to meet the unique market and regulatory needs of the permissioned financial data access ecosystem in North America. As we do that, we’re tracking other emerging standards and protocols and looking for opportunities to align or reference where it makes sense.

In this work, the industry has an opportunity to build on the great work of the past, support innovation, and deliver a great “front door” for data sharing. At the heart of it all is the consumer; as they seek to use a new generation of tools, the industry shares a responsibility to help them do so securely. We welcome financial institutions, AI tool providers, and others to help build that together.

Notes

  1. APIs come in many shapes. In this context, we use the term “purpose-built API for third-party data sharing” to refer to a secure interface designed for standardized data sharing between companies, incorporating tokenized credentials, structured requests, and granular authorization. Some companies today are building agent-protocol layers (e.g., MCP, A2A) that enable discovery and communication for AI agents and tools; at their heart, these channels are typically still anchored in APIs.
  2. Here we use the term “screen-scraping” broadly to refer to any access method in which a data-requesting company or tool uses the consumer’s credentials to log in to a data provider’s “human” channels, such as its website, and retrieve data from there. Under this broad definition, screen-scraping can take many forms, including HTML parsing and reverse engineering of a bank’s web and mobile APIs. It can come from a developer-written script, an AI-crafted script, or an AI tool or agent operating a browser in real time (with or without human supervision).
  3. While there are many flavors of screen-scraping and different ways to build such tools, credential-based access to financial institution websites largely fails to deliver this full set of features.