Capital One joins banks in calling for rules on AI shopping agents before fraud takes off

By Zendoric · September 26, 2026.

Capital One Financial (COF), the US digital bank known for its credit cards and with a market value of around $122.8 billion according to Simply Wall St data, has joined an international coalition of banks warning about the risks of AI-powered "shopping agents." The coalition is calling on regulators for clearer safety and disclosure rules for these automated purchasing tools, as the article itself reports. No names of the other institutions are cited, nor any specific legislative proposal: for now this is an industry warning, not a rule.

The underlying argument is simple. An AI agent that accesses a user's payment data and completes purchases autonomously multiplies the opportunities for scams, mistaken orders or unauthorized use of a stored card. For Capital One that is not an abstract risk: it concentrates exposure precisely in its digital banking and card platforms, where fraud, chargebacks (charge reversals the bank has to absorb) and remediation costs hit the bottom line directly.

Simply Wall St frames the news within its own investment thesis on the stock: Capital One is already betting heavily on technology and AI to bring down fraud costs and sustain the quality of its credit portfolio, while the integration of Discover -the payments network it absorbed in 2025- gives it scale against rivals such as American Express or Synchrony. The warning about shopping agents, the outlet notes, underscores the risk side of that same bet: more regulatory scrutiny may translate into higher compliance spending, or into penalties if controls are judged insufficient. As signals to watch, the analysis points to the trajectory of fraud and delinquency relative to competitors, and to any formal regulatory guidance or investigation into AI-driven transactions.

As sector context, 2025 and 2026 have been the years in which "agentic commerce" went from promise to product: major payment networks and AI platforms have begun deploying protocols so that an assistant can search, compare and complete a purchase without the user typing in a card at every step. The logic is always the same, whoever is applying it: to be useful, the agent needs permission to spend someone else's money without supervision at every click. And that permission is exactly what any attacker would want to steal.

That is what makes this story interesting, beyond the fraud headline. The ones pushing for regulation are not the ones building these shopping agents, but the ones who pay the bill when they fail: banks do not make the assistants, they absorb the fraud, the chargeback and the cost of making the defrauded customer whole. It makes sense that it is the banking industry, and not the AI platforms, that is first to demand transparency rules about who is liable when an agent buys badly. It is the same dynamic we already saw with fraud industrialized by autonomous agents on other fronts: the autonomy that makes the tool useful is the same autonomy that widens the attack surface.

Our reading is that warnings of this kind, however generic, mark the start of a necessary phase: building the infrastructure of trust before convenience takes over unchecked. In the short term, every new layer of autonomy -an agent that buys, that pays, that decides for you- will grow faster than the identity verification mechanisms, auditable spending limits and traceability that would make it safe by design; that means more fraud and more stolen identities before the safeguards arrive, not fewer. In the long term, however, well-governed agentic commerce -with the agent identified, spending capped and every transaction traceable- would be one of the most tangible pieces of the abundance AI promises: buying, paying and managing money without friction, freeing up people's time and attention for what really matters. The question this banking warning leaves behind is not whether agentic commerce is coming, but whether trust gets built in time or is dragged along behind the frauds that will force it to be built.