Artificial intelligence is rapidly moving from helping consumers make purchasing decisions to making those decisions for them. So-called “agentic AI” or “AI shopping agents” could search for financial products, compare prices and terms, negotiate with providers, open or close accounts, switch providers and complete transactions without the consumer participating in each step.

In the Consumer Finance Monitor podcast we are releasing today, Alan Kaplinsky (founder, former chair for 25 years and now Senior Counsel of the Consumer Financial Services Group at our firm spoke with Professor Mark Bartholomew of the University at Buffalo School of Law, co-author with Professor Samuel Becher of “The End of Shopping,” a forthcoming article in the William & Mary Law Review. The article explores how autonomous shopping agents could transform consumer commerce and raises fundamental questions concerning consumer autonomy, privacy, competition and consumer protection.

The implications for financial services are especially significant. Consumers could use AI agents to select credit cards, auto loans, mortgages, insurance policies, bank accounts, certificates of deposit and investment products. An agent potentially could compare thousands of offerings in seconds and execute a transaction that a consumer might otherwise never undertake because of the time and complexity involved.

For banks, lenders, insurers, credit-card issuers and investment firms, agentic AI could reduce customer-acquisition costs, increase switching and intensify competition. But it also could make technology companies the principal gatekeepers between financial institutions and their customers. The company controlling the agent could determine which products consumers see, how those products are compared and whether a particular provider receives the consumer’s business.

Professor Bartholomew argues that policymakers should address these issues before agentic commerce becomes entrenched. Among the potential guardrails he discusses are independent audits, “algorithmic nutrition labels,” data portability, protections against self-dealing, meaningful consumer control—including an “off” switch—and periodic review or sunset provisions for new regulation.

The central question for financial services providers and regulators is whether the existing consumer-protection framework will work if AI agents become the principal interface through which consumers select and purchase financial products.

This episode builds on our earlier March 12, 2026 podcast with Professor Oren Bar-Gill of NYU Law School discussing Agentic AI in Consumer Financial Services: Opportunities, Risks and Emerging Legal Frameworks, our May 21, 2026 podcast with Professor Mark Geistfeld discussing the American Law Institute’s Principles of the Law, Civil Liability for Artificial Intelligence. Together, these three programs examine the tort law, contract law and other legal challenges posed by increasingly autonomous AI systems.

From Financial Shopping to Financial Decisions Made for Us

For decades, technology has made financial shopping increasingly convenient. Consumers moved from bank branches and paper applications to online banking, mobile apps and digital comparison tools. They can now receive personalized recommendations, prequalified offers and automated investment advice.

Agentic AI could represent the next—and much more consequential—step.  There is an important distinction between an AI assistant and an AI shopping agent. Today, a consumer can ask an AI assistant to recommend a credit card, mortgage, insurance policy or investment product. The consumer still evaluates the recommendation, makes the decision and completes the transaction.

An agentic system potentially could do all of that itself.  A consumer might tell an AI agent: “Find me the lowest-cost auto insurance policy that provides the coverage I need.” Or: “Find me the best one-year CD with FDIC insurance.”  Or: “Find me a credit card with no annual fee, a low introductory rate and rewards that fit my spending.”  The agent could search available alternatives, compare prices and terms, make a recommendation—or make the decision pursuant to instructions the consumer previously provided—and complete the transaction.  It also could monitor the consumer’s accounts and act later. An agent might move funds when a CD matures, refinance a loan when rates fall, switch insurance carriers after a premium increase or cancel a credit card that no longer offers competitive terms.

For financial institutions, this could change the customer relationship fundamentally. The consumer may no longer begin with a bank’s website, a lender’s advertisement or an insurer’s agent. The consumer may begin with an AI platform that decides which providers and products deserve consideration.

Why Financial Services May Be an Especially Attractive Use Case

The financial services industry may be one of the areas in which consumers are most willing to delegate shopping decisions to AI.

Comparing credit cards, deposit accounts, auto loans, mortgages, insurance policies and investment products can be considerably more complicated than comparing consumer goods. Terms and conditions may be lengthy and difficult to compare. Pricing may depend on individualized underwriting. Consumers frequently lack the time or expertise to evaluate competing products.

An AI agent could theoretically examine hundreds or thousands of offerings, compare rates, fees, coverage, eligibility requirements and other terms, and identify products that best fit a consumer’s stated preferences.  The potential benefits could be substantial.

Consumers frequently remain with their existing bank, insurer, lender or credit-card issuer not because it offers the best product but because switching is difficult. An agent that could handle the process of comparing alternatives and switching providers could dramatically reduce those switching costs.

For example, an agent could identify a higher-yield deposit account, transfer funds, update direct-deposit instructions and close the old account. It could compare mortgage-refinancing offers, gather documents and submit applications. It could evaluate insurance premiums and coverage, select a new carrier and cancel the existing policy. It could compare credit-card rewards and fees and recommend a replacement card.  That could increase competition and pressure financial institutions to offer better prices and service.

But financial-services providers should recognize that the party making the initial decision about which products consumers see may no longer be the consumer or even a human intermediary. It may be an AI agent operating according to an algorithm supplied by another company.  That creates an entirely new potential gatekeeper.

Implications for Banks, Lenders, Insurers and Investment Firms

Agentic AI could affect nearly every stage of the financial services relationship.

Banks may face more frequent movement of deposits as agents compare interest rates, fees, account features and convenience. A bank’s ability to retain customers through inertia or the difficulty of switching could diminish. At the same time, banks may gain access to new customers if their products perform well under objective comparisons.

Lenders may encounter a more efficient but more demanding marketplace. An AI agent could compare mortgage, auto-loan and personal-loan offers based on annual percentage rates, fees, repayment terms, underwriting requirements and closing costs. Lenders may need to make pricing and eligibility information more accessible to automated systems and may face greater scrutiny if their offers are difficult for agents to evaluate.

Insurers could see similar changes. Agents might compare premiums, deductibles, exclusions, coverage limits, claims practices and renewal terms. Insurers may need to explain complex coverage in machine-readable formats while ensuring that automated comparisons do not omit material limitations.

Credit-card issuers could face rapid customer switching based on annual fees, interest rates, rewards, introductory offers, balance-transfer terms and penalty provisions. Issuers also may seek to influence the agents that determine which cards consumers see. That could create questions about referral payments, sponsored placement and whether an agent’s recommendation is genuinely based on consumer benefit.

Investment firms may confront particularly sensitive issues. An agent could select mutual funds, exchange-traded funds, brokerage accounts, retirement products or advisory services. The agent’s recommendations could affect asset allocation, risk exposure, fees and long-term financial outcomes. Investment firms and regulators will need to consider how existing suitability, best-interest and fiduciary principles apply when an AI system makes or implements recommendations.

Across all sectors, financial institutions may need to compete not only for consumers but also for favorable treatment by the platforms controlling the agents.

Who Is Making the Decision?

Suppose a consumer instructs an AI agent to find the best mortgage available, but the agent submits an application to a lender with a higher rate because that lender pays the platform a referral fee. Is the consumer bound by the application? Is the AI platform responsible? Is the lender responsible for accepting business generated through a conflicted recommendation?

Professor Bartholomew views this in part as a delegation problem. The consumer has delegated authority to the AI system, but traditional agency law may not provide a complete answer. Traditional agency doctrine developed around relationships involving human principals and human agents. An AI system does not have the same incentives, motivations or ability to respond to liability that a human agent does. The company operating the system, the company developing the underlying model and the financial institution receiving the business may each play a role, but existing law may not clearly allocate responsibility among them.

The issue is especially important in financial services because transactions can create long-term obligations. A mistaken purchase of a consumer product may be inconvenient. A mistaken mortgage, insurance policy, investment or credit-card application can affect a consumer’s finances, credit history and legal rights for years.

Financial institutions therefore may need to determine what authority an AI agent has, how that authority is verified and when a transaction requires meaningful consumer confirmation.

Conflicts of Interest and Self-Dealing

An agent supposedly working for the consumer could receive referral fees, advertising payments, data benefits or other compensation from a bank, lender, insurer, credit-card issuer or investment firm. It might then have an incentive to recommend that provider even though another provider offers a better product for the consumer.

The conflict could take several forms:

  • A credit-card issuer could compensate an agent for completed applications.
  • An insurer could provide data or other benefits in exchange for favorable treatment.
  • A lender could pay for leads generated by an agent.
  • An investment firm could compensate an agent for directing assets into particular products.
  • A platform could favor affiliated financial institutions over unaffiliated competitors.

These arrangements could resemble familiar forms of advertising or referral compensation, but agentic AI may make the conflict more difficult for consumers to detect. The agent may not display a conventional advertisement. It may simply omit certain products, rank one provider more favorably or describe competing products less prominently.

That raises a fundamental question: Is the AI agent actually working for the consumer?

For financial regulators, the issue may require more than disclosure. If an agent’s incentives are fundamentally inconsistent with the consumer’s interests, a disclosure may not be sufficient. Regulators may need to consider restrictions on compensation arrangements, requirements for independent comparison or duties requiring the agent to act in the consumer’s best interest.

Consumer Protection Law May Have to Change

Much of consumer protection law assumes that a human consumer is on the other side of the transaction.

Disclosures are provided so that consumers can read and understand them. Advertising laws are designed in part to prevent businesses from exploiting consumers’ cognitive biases. Regulation often focuses on whether consumers have received sufficient information to make an informed decision.

Financial-services laws similarly rely on consumer interaction with disclosures, applications, notices and explanations. Truth-in-lending disclosures, deposit-account disclosures, insurance documents, privacy notices and investment disclosures are generally designed for human review.

Agentic AI changes that model.  An AI agent does not become confused because a price ends in “.99.” It does not necessarily respond to an emotionally manipulative advertisement. It can theoretically process far more information than an individual consumer.

But that does not mean the agent is invulnerable.  The relevant vulnerabilities may instead be algorithmic.  An agent could be programmed to favor a particular financial institution. It could be manipulated by a business seeking to influence how its products are presented to AI systems. It could fail to search a sufficiently broad universe of alternatives. It could misunderstand a consumer’s financial objectives or risk tolerance. Or its operator could use the enormous amount of personal information accumulated about the consumer to steer purchasing decisions in ways the consumer does not understand.

The focus of consumer protection therefore may need to shift from protecting the human consumer’s cognitive processes to protecting the integrity of the algorithm making decisions on the consumer’s behalf.

Regulators also may need to reconsider what it means for a consumer to receive a disclosure. If an AI agent reads and processes a disclosure but the consumer never sees it, has the purpose of the disclosure requirement been satisfied? If the agent summarizes a complex loan term incorrectly, who is responsible? If the agent rejects a product because it cannot interpret the provider’s disclosures, does that create an unfair competitive advantage for institutions with more machine-readable documentation?

Fair Lending, Insurance and Investment Concerns

Agentic AI could create new risks under existing financial services laws.

For lenders, an agent’s search and recommendation process could affect which consumers apply for credit and which lenders receive applications. If an agent systematically excludes certain lenders or products, the effects could have implications under fair-lending and fair-access principles.

The agent itself also could use sensitive information in ways that influence recommendations. A consumer’s location, financial history, family circumstances or purchasing patterns could affect which products the agent presents. Even if the agent does not use a protected characteristic directly, its use of related information could produce discriminatory outcomes.

Insurers may face similar concerns. An agent could compare policies using individualized pricing, claims history, location and other data. Regulators may need to examine whether automated recommendations reinforce unfair distinctions or obscure the factors driving coverage and pricing decisions.

Investment firms face additional concerns involving risk tolerance, suitability, best-interest obligations and the possibility that an agent will favor products generating higher compensation. An agent that automatically reallocates a consumer’s portfolio could make decisions that are technically consistent with historical preferences but inappropriate in light of changed circumstances.

These issues suggest that financial regulators may need to examine not only the conduct of financial institutions but also the design and operation of the AI platforms that influence financial decisions.

The Need for Guardrails

Professor Bartholomew’s principal concern is not simply determining liability after an AI agent causes harm. He emphasizes the need for ex ante safeguards that make it more likely that the system will operate in the consumer’s interest from the beginning.

One proposal is independent auditing.  AI shopping agents could be tested against standardized benchmarks to determine, for example, how broadly they search, whether they consistently identify competitive rates and terms, whether they accurately represent fees and limitations, whether they favor particular financial institutions and whether their recommendations vary based on protected or sensitive characteristics.

For financial regulators, audits could provide a way to evaluate agentic systems without requiring regulators to review every individual transaction. Audits also could help institutions demonstrate that their systems are not designed to manipulate consumers or evade existing legal requirements.

Another proposal is an “algorithmic nutrition label.”  Just as consumers can look at a nutrition label to evaluate food without independently testing its contents, consumers could receive standardized information about an AI shopping agent. Such information might disclose:

  • How many banks, lenders, insurers or investment firms the agent typically searches.
  • Whether the agent includes affiliated or paying providers.
  • How the agent is compensated.
  • Whether the agent receives referral fees or other benefits.
  • How the agent performs against benchmark rates, fees and terms.
  • What consumer data the agent uses.
  • Whether the agent can open, close or transfer accounts without additional approval.
  • How consumers can challenge or reverse an agent’s decision.

The objective would be to give consumers and regulators meaningful information with which to evaluate competing AI agents.

Competition and Data Portability

Agentic AI also raises significant competition concerns.   The company controlling an AI shopping agent could become the gatekeeper through which consumers access an enormous portion of the financial marketplace. That could give already dominant technology companies an even greater competitive advantage.

A platform that controls the consumer interface could influence which banks, lenders, insurers, credit-card issuers and investment firms receive business. It also could use transaction data to develop competing financial products or negotiate more favorable terms for affiliated providers.

Financial institutions may therefore need to consider whether access to consumers is becoming dependent on a small number of technology platforms. Regulators may need to examine exclusive arrangements, preferred placement, interoperability and the possibility that dominant platforms will discriminate against smaller financial institutions.

Data portability could be particularly important. Suppose a consumer has used one AI agent for years. The agent has accumulated extensive information about the consumer’s preferences, purchasing history, financial circumstances, risk tolerance and account relationships. If the consumer wants to switch to a competing agent, the consumer should not necessarily have to start over.

Professor Bartholomew therefore advocates data portability so that consumers can take their information with them when they switch providers.  Portability could also promote competition among financial institutions. If consumers can easily transfer account histories, preferences and authorization information, they may be more willing to switch banks, lenders, insurers or investment firms.  Without portability, the companies that establish an early lead in agentic commerce could make it extremely difficult for consumers to leave.

Privacy and Cybersecurity Risks Will Increase

The benefits of agentic AI depend heavily on access to information about the consumer.  The more an agent knows about a consumer’s purchasing history, financial circumstances, preferences and personal life, the better it potentially can tailor its decisions.

But that same information creates substantial privacy and cybersecurity risks.  An AI shopping agent could become a comprehensive repository of information about virtually everything a consumer buys, considers buying or can afford to buy. It could know the consumer’s income, debts, savings, insurance coverage, investment objectives, medical expenses and financial vulnerabilities.  The aggregation of that information could make an agent particularly attractive to hackers, data brokers and businesses seeking to influence consumers.

Banks, lenders, insurers, credit-card issuers and investment firms will need to determine what information they share with agents, how they authenticate an agent’s authority and how they prevent unauthorized transactions. They also may need to distinguish between information necessary to complete a transaction and information that could be used to profile or manipulate a consumer.

Financial regulators may need to address whether existing privacy and cybersecurity requirements adequately cover data held by AI platforms that are not themselves traditional financial institutions.

The challenge will be finding the right balance between allowing AI agents to access enough information to provide meaningful benefits and preventing that information from being exploited.

Consumers Need an Off Switch

One of the simplest proposals discussed in the podcast is also one of the most important: Consumers should be able to turn the agent off.

There may be many financial decisions that consumers want an AI system to handle automatically. A consumer may want an agent to move idle cash into a higher-yield account, monitor insurance premiums or identify lower-cost refinancing opportunities.

But there are other decisions where consumers want to participate directly.  A consumer may want to approve a mortgage application, select an investment strategy, purchase life insurance or close a long-standing bank account personally. Consumers also may want to prevent an agent from acting during periods of financial stress, illness or unusual market volatility.  Meaningful consumer control should include clear limits on the agent’s authority, advance notice of significant transactions, the ability to require human approval and a practical way to reverse or challenge decisions.

Professor Bartholomew also suggests features such as “explore” or “surprise” modes that would prevent an agent from relying exclusively on a consumer’s historical preferences. In financial services, that could mean requiring an agent to consider new providers, alternative products or different investment approaches rather than continually recommending the same institutions and products.  Otherwise, the very personalization that makes agentic AI useful could become a mechanism that continually reinforces past behavior.

What Should Financial Institutions Do Now?

Banks, lenders, insurers, credit-card issuers and investment firms should not wait for regulators to resolve every legal question before preparing for agentic commerce.

They should consider whether their product information is accurate, complete and machine-readable. Rates, fees, eligibility requirements, exclusions, rewards terms, risk disclosures and other material information should be presented in formats that AI systems can interpret without losing important context.

Institutions also should review relationships with AI platforms. Referral fees, preferred placement, sponsored recommendations and data-sharing arrangements may create conflicts that are difficult to explain to consumers. Firms should be prepared to demonstrate that their arrangements do not result in deceptive steering or unfair treatment.

Governance programs should address:

  • How the institution authenticates AI agents.
  • What authority an agent has to open, close or modify an account.
  • When human confirmation is required.
  • How the institution records and explains agent-generated transactions.
  • How errors are corrected.
  • How complaints involving AI agents are investigated.
  • How the institution monitors disparate impacts.
  • How third-party AI platforms are supervised.
  • How consumer data is protected and deleted.

Institutions also should consider whether their existing customer-service models are adequate when the customer’s first interaction is with an AI agent rather than a human consumer.

Do We Need a New AI Regulator?

The Federal Trade Commission clearly has a role to play in policing deceptive claims and other misconduct involving AI shopping agents. But Professor Bartholomew questions whether existing agencies are equipped to address all of the issues raised by agentic commerce.

Financial regulators—including the Consumer Financial Protection Bureau, federal banking agencies, state banking and insurance regulators, the Securities and Exchange Commission and the Financial Industry Regulatory Authority—also may have important roles.

Some issues involve technical standards, interoperability, data portability and auditing. Others involve potentially complex questions concerning fiduciary duties, self-dealing and the relationship between consumers and the companies operating AI agents.

Regulators will need to determine whether existing laws can be applied to AI platforms that influence financial decisions but are not themselves banks, lenders, insurers, broker-dealers or investment advisers. They also will need to determine how responsibility should be allocated when a financial institution relies on an agent operated by a third party.

Congress may ultimately need to establish a broader framework, potentially including a specialized regulatory body or assigning additional responsibilities to existing agencies.

There is also an important institutional question: How should AI regulation be structured so that the rules do not change dramatically with every change in administration?

Whatever regulatory structure emerges, the financial-services industry and technology companies may benefit from having relatively stable rules of the road. Banks, lenders, insurers, credit-card issuers and investment firms need to know what standards will apply as they make investments in agentic commerce.

Don’t Lock in the Wrong Rules

One of the more unusual recommendations in “The End of Shopping” is that some AI regulations should contain sunset provisions.

The rationale is straightforward. We do not yet know exactly how agentic commerce will develop or whether particular regulatory approaches will work as intended.

Rather than adopting rules that remain in place indefinitely, policymakers could require the rules to be reconsidered after a specified period, perhaps three or five years.  That would allow regulators and Congress to evaluate what actually happened, identify unintended consequences and modify the regulatory framework as the technology develops.

For financial regulators, periodic review could be especially useful. Regulators could assess whether agents are increasing competition or merely shifting market power to technology platforms; whether consumers are receiving better rates and terms; whether automated recommendations are producing discriminatory outcomes; and whether disclosures and consent mechanisms are meaningful in practice.

It is a particularly sensible concept for a technology whose capabilities may change substantially within a few years.

What Does the Future Hold?

It is difficult to predict how quickly consumers will embrace agentic shopping. But it seems increasingly likely that AI systems will move beyond recommending products and services and begin performing transactions on consumers’ behalf.  If that happens, the implications for consumer financial services could be enormous.

An AI agent could become the consumer’s principal interface with banks, credit-card issuers, lenders, insurers and investment firms. A consumer might no longer visit a bank’s website, compare credit cards or call competing mortgage lenders. The consumer might simply tell an AI agent what outcome is desired and allow the agent to determine how to achieve it.  That could reduce search costs, increase competition and make it substantially easier for consumers to switch providers.  It also could change how financial institutions compete. Product pricing, transparency, machine-readable information and the ability to integrate with AI platforms may become as important as branch networks, advertising and brand recognition.

But agentic AI could also create a new class of gatekeepers with unprecedented access to consumer information and unprecedented ability to influence financial decisions.  The central policy question therefore should not be whether agentic AI is good or bad. It is how to ensure that the technology actually serves the consumer whose interests it purports to represent.

That will require regulators, courts, businesses and policymakers to rethink some fundamental assumptions underlying consumer protection law. It also will require financial institutions to reconsider how they market products, obtain consent, manage third-party relationships and demonstrate compliance.

For decades, the law has assumed that the consumer is the person making the purchase.

The era of agentic commerce may force us to confront a very different question:

What happens when the consumer no longer does the financial shopping?

Professor Bartholomew’s article, “The End of Shopping,” provides a provocative starting point for answering that question.