Today's consumer protection laws assume a human is shopping — what happens when an AI agent clicks 'buy' for you?

This explores whether today's consumer protection rules, written for people clicking through websites, still hold up when an AI agent shops and buys on your behalf. The corpus mostly says they don't. The interesting part is that the failure goes deeper than a few missing rules: the law assumes things about who the consumer is and how a purchase happens, and agents break those assumptions. EU consumer law is built around an 'average consumer' with human attention, human biases and human vulnerabilities, and its disclosure rules assume someone actually reads the disclosure. When the buyer is an algorithm, those tests stop making sense (Can EU consumer law protect AI agents as algorithmic buyers?). There is also a mechanical problem. Many protections live in interface design: a mandatory 'order with obligation to pay' button, a withdrawal form, a clearly placed price. When agents deal with merchants directly through APIs, there is no screen for those rules to attach to. Busch's proposed fix is to move the obligations down to the protocol layer, for example by requiring a 'withdrawal API' so the right to cancel survives even when no human ever sees a checkout page (Can EU consumer law rules survive agent-to-agent API transactions?).

A second line of thinking asks not 'how do we protect the consumer?' but 'what does the agent owe the consumer?' Owens and Law argue that shopping agents should carry the same fiduciary duties as human brokers and agents, meaning loyalty, care and no secret self-dealing. They also argue this has to happen soon, because Google, OpenAI and Robinhood are scaling agent features faster than norms can form (Should AI shopping agents face fiduciary duties like human brokers?). Senator Warner's draft AI AGENT Act combines both approaches. It would impose a 'best-interest' duty on agents, create an FTC registry of trusted agents, and require large platforms to let those agents in (Could federal AI agent registries solve platform access unfairness?). The access piece matters because, without rules, platforms simply block agents they don't like through their terms of service. Amazon's block of Meta's Muse agent is the example, and it shows there is no standard way for an agent to prove who it is and who authorized it (What identity standards do shopping agents actually need?).

The less obvious finding is that the biggest harms may not come from agents betraying their users. In Microsoft's Magentic Marketplace experiments, most of the lost value came from agents making mistakes in complicated negotiations, not from agents secretly serving someone else (Do agent mistakes or misaligned incentives cause marketplace welfare loss?). Fiduciary duties target disloyalty, so a duty of competence may matter as much. A second, subtler harm comes from the market around the agent. One simulation found that when a platform tunes its rankings and sellers tune their product descriptions to exploit the same agent biases, consumer harm more than doubles, with no coordination between them (Can platform ranking and seller descriptions separately harm AI shoppers?). Unfair-practice law usually looks for one bad actor. Here the harm comes from two parties each acting legally on their own, which existing doctrine is poorly equipped to catch.

Liability is also moving in ways consumer law doesn't track. As agents take over checkout, the behavioral signals merchants use to spot fraud, such as typing rhythm and mouse movement, disappear. Yet under every proposed agent-payment protocol the merchant stays legally responsible as the merchant of record (Does agentic commerce shift fraud liability onto merchants?). Market structure feeds into this too. Whether agentic commerce settles into open systems (any agent can shop anywhere) or closed ones (proprietary agents only) will shape how much bargaining power consumers' agents actually have (Will agentic commerce platforms adopt open or closed models?). Intermediaries also aren't vanishing the way some predicted. In travel, online travel agencies are becoming the backend infrastructure agents book through, which leaves familiar middlemen in place inside a less visible layer (Are AI agents actually removing online travel agencies from bookings?).

The corpus is clearest on diagnosing the gaps and thinner on evidence that any fix works. The proposals (protocol-layer obligations, fiduciary duties, registries) are arguments and drafts, not tested regimes. The thing worth taking away: protecting consumers in agentic commerce may depend less on rewriting disclosure rules and more on regulating the agent's competence and the APIs it talks to.