Regulators should act now to control the potential harms of AI shopping agents.
Imagine waking up to find that, while you slept, an artificial intelligence (AI) shopping agent switched your auto insurer, rebooked your vacation at a lower price, and ordered running shoes matched to your stride and preferred terrain. That prospect no longer belongs to science fiction. AI companies, payment networks, and major retailers—including Walmart, Amazon, and Target—already offer shopping research tools and AI agent modes, and are racing to build tools that go beyond product recommendations toward automated transactions.
AI agents hold real promise for consumers. These agents can scan billions of product listings, track price histories, parse impenetrable terms of service, and synthesize thousands of reviews—all while avoiding the cognitive limitations and behavioral biases that often lead human shoppers astray. In a world of fake reviews, dark patterns, and choice overload, a competent digital agent could genuinely help consumers make better choices.
But regulators should not confuse convenience and efficiency with harmlessness.
Our entire regulatory apparatus for consumer protection assumes that a human is doing the shopping. False advertising statutes protect against deceptive claims. Disclosure mandates, such as truth-in-lending and nutritional labeling, put information in front of human eyes. Trademark law makes it easier for consumers to identify the source of a product. These frameworks share a common premise: Somewhere in the transaction, a person is deliberating.
When an AI agent shops on our behalf, that premise may collapse. As we argue in a forthcoming article, shopping agents will not simply help consumers navigate markets; they will increasingly govern how markets work and what shoppers choose. Once an AI tool decides what products to show, which sellers to prioritize, and when to execute a purchase, it becomes a form of market infrastructure.
If leading AI agents are controlled by dominant retail platforms, the risk is that these platforms will self-preference, creating a new form of corporate power. A biased algorithm, masquerading as neutral optimization, may be much harder to detect and contest than a biased salesperson.
The risks are not only competitive or informational, but also personal. Much shopping is mundane and transactional, but we often still exercise judgment, form tastes, express values, and discover wants in our shopping decisions. We construct and signal who we are through what we choose to buy or to avoid. We do this, for example, when we choose fair-trade and environmentally friendly products or avoid products associated with labor abuses or political causes we oppose. Even the period between selecting a product and receiving it can generate pleasure and anticipation, reinforcing our sense of agency.
A shopping agent trained on a consumer’s past shopping behavior risks becoming a machine for freezing the self, ossifying consumer identity into an algorithmic profile. It also compresses anticipation into an instant notification, while narrowing exposure to the unexpected.
Shopping is also communal. We select gifts that signal care, browse alongside others, and demonstrate thrift as a shared virtue. When algorithms replace these acts, the relational meaning is hollowed out. The result may be more efficient or optimal, but the human gesture disappears.
More broadly, navigating the marketplace cultivates habits of comparison, thrift, and critical evaluation that carry beyond commerce into civic life. Outsourcing all of that to machines risks eroding capacities we cannot easily rebuild. What looks like frictionless convenience may quietly strip the marketplace of the qualities that make participation in it meaningful.
Regulators should act now, before the architecture of AI shopping hardens around platform incentives.
First, the law could mandate “algorithmic nutrition labels” in the form of standardized smart disclosures. Such disclosures could reveal which sources agents consult, whether they favor particular affiliated sellers, and how recommendations are weighted. These disclosures should be verified by third-party auditors, available for public scrutiny, and presented in a user-friendly format.
Second, regulators could require meaningful data portability so that consumers can transfer their preferences, histories, and agent profiles across competing firms, preventing consumers from being locked into any one service. Portability should be viewed not just as a tool to promote competition but also as a means to protect autonomy, giving users the ability to exit ecosystems that no longer serve them.
Finally, every shopping agent should have a real “off button.” This should be a user-facing mechanism that allows consumers to disable automation when purchases carry relational, cultural, or political significance. That may be the case, for example, with giving a gift or participating in an economic boycott.
Although these are possible starting points, the technological ecosystem moves fast, and the market implications are difficult to predict. Any legislation governing AI shopping agents should include sunset clauses, allowing rules to expire automatically after a set period unless reauthorized. Regulatory frameworks must evolve with our technological tools.
The broader imperative here is that legal frameworks should not wait for harms to crystallize before responding to technological disruption. Too often, regulation arrives only after technologies have become embedded in everyday life and path dependencies have hardened. Remediation at that stage is costly, politically fraught, and likely to be blocked by the very industries that benefited from the delay. The proposals we outline here aim to shape the architecture of AI shopping as it is being built, not after it has become entrenched.
Consumer spending accounts for roughly two-thirds of American gross domestic product, and the habits of evaluation and judgment that it cultivates permeate our societal fabric. When software begins to choose, rank, and buy on our behalf, the law is no longer regulating a convenience feature. The question is who controls consumption, how markets are structured, where power lies, and how much agency consumers retain in everyday life.
AI can now shop for you. Now the law must adapt before private platforms turn automated shopping into the new infrastructure of consumer life.
