Fraud in the age of agentic: who bears the risk?

Jeff Otto, Chief Marketing Officer at Riskified, discusses the new age of agentic commerce, the risks associated with it, including new types of fraud, and how to streamline this technology further.

As AI shopping agents gain traction, merchants face rising fraud risks and unresolved questions over who ultimately pays the price.

The theoretical future of agentic commerce became a tangible reality in 2025. With Apple integrating advanced, agent-like intelligence into Siri for hundreds of millions of users and Google embedding its ‘AI Overviews’ directly into search, AI-powered shopping tools are no longer just advisors; they are rapidly becoming autonomous actors on behalf of consumers. And with the latest launch of Google’s Agent Payments Protocol to enable agent transactions, the agentic shift is poised to redefine retail, bringing both incredible opportunities and profound risks.

But every historical change in shopping behaviour has triggered a parallel wave of fraud and abuse. Agent-driven commerce will be no different. The velocity and scale at which AI agents operate mean that fraud risks will intensify faster than ever before, raising a fundamental question –  when an AI agent commits fraud, who ultimately pays? Is it the merchants, the card issuers, or the consumers themselves?

The new commerce funnel: from discovery to execution

• The early stage of agentic commerce is already here. Consumers are using multimodal models like OpenAI’s GPT-4o and Google’s Gemini to discover products, find discounts, and skip directly to checkout. This seemingly small shift carries outsized consequences.
• Disputes and friendly fraud will rise: as agents summarise product details, consumers will be more likely to miss fine print on shipping conditions or return policies, leading to a rise in abusive returns, refunds, and non-fraud chargebacks.
• Pricing competition will accelerate: agents can scour thousands of sites for the lowest price, exposing merchants to aggressive price arbitrage.
• Identity signals will vanish: agent-led purchases often default to guest checkouts, eroding the account-based behavioural signals, like device fingerprints and session tracking, that merchants rely on to fight fraud.

Next-generation tools, from Amazon’s ‘Buy for Me’ to the integrated agents in Apple Intelligence, will go even further, completing purchases autonomously. While adoption is low now, the convenience is undeniable, and the trajectory is clear. We are heading to a more extreme version of the liability problem seen when online travel agencies reshaped airlines into inventory suppliers, leaving fraud disputes unresolved. An example that underscores how the disintermediation of who’s making the decision to transact creates systemic risk.

A glimmer of hope: the collaborative ecosystem takes shape

The future of agent-driven shopping could take several forms, but a recent announcement from Google offers a strong indication of where the industry is heading. In September 2025, Google, along with over 60 partners, including major payment networks like Mastercard and American Express and payment processors like PayPal, announced the Agent Payments Protocol (AP2). This open protocol is designed to create a common language for secure, compliant transactions between AI agents and merchants.

This development is a direct answer to the industry's concerns about fraud and liability in agentic commerce and offers a compelling point of view on the potential futures of this new landscape:

1. The walled garden vs. the open protocol: while some players are developing proprietary agents within their own ecosystems, the broader industry is pushing back against a purely ‘walled garden’ approach. Open protocols encourage interoperability and prevent a fragmented ecosystem where merchants would have to support dozens of different agent technologies.
2. A move away from disintermediation: the most disruptive and dangerous future for merchants is one where they are completely disintermediated by powerful super-agents. Collaborative frameworks, by design, are an attempt to prevent this. By creating shared standards, they give merchants and payment providers a seat at the table, allowing them to participate in the agentic commerce ecosystem in a secure and standardised way. They provide a mechanism for verifying both the consumer and the agent, which is a critical step in mitigating fraud.
3. The rise of the ‘collaborative ecosystem’: the push for open standards is a massive step towards the ‘collaborative ecosystem’ model. It provides the technical foundation for a future where merchants and third-party agents can interoperate through secure APIs. While liability will still need to be negotiated, these protocols provide the rails upon which those agreements can be built. It’s a clear signal that the industry's major players recognise the need for shared standards to ensure security and trust.

The escalating fraud challenge: an identity crisis

Even with promising developments like Google AP2, the fraud challenges are intensifying. Early data from the Riskified merchant network shows that traffic referred from LLMs is already riskier in key industries. For example, LLM-referred traffic from a large ticketing merchant was 2.3x riskier compared to Google search traffic.  In another example, an electronics merchant showed 1.8x riskier traffic.

The core issue is a new identity crisis. Today’s fraud prevention depends on signals that prove a human is behind the screen. When an agent makes a purchase, those signals vanish, leaving merchants exposed. This automated traffic, even when legitimate, often mimics bot activity, increasing the risk of false declines that frustrate genuine customers and erode trust.

Meanwhile, fraudsters are weaponising agents to scale their attacks. They can automate credential stuffing, phishing, and mass order execution to overwhelm merchant defences. The biggest new threat might be ‘Compromised Agent-as-a-Service’. A single hijacked user profile connected to an agent with deep permissions, providing access to email, calendars, and payment credentials, can trigger fraudulent purchases across dozens of stores, multiplying disputes and making liability nearly impossible to trace. This isn't just a stolen credit card; it's a stolen digital identity with autonomous purchasing power.

Taking action: building resilience now

Agentic commerce is already here, and it is exposing critical gaps in traditional fraud models. The question of who pays for fraud in this new world will define the next era of digital commerce. The release of new MCPs (Model Context Protocol) is a sign that the industry is moving in the right direction, but they are just the beginning.

Durable solutions will require collective action. Technology alone is not enough. Merchants, issuers, networks, and regulators must collaborate to build on these emerging standards before disputes cascade unchecked. This includes developing new protocols for Know Your Agent (KYA), a necessary evolution of KYC designed to credential the agents themselves.

Without this cooperation, the burden of fraud will fall unevenly, leaving merchants disproportionately exposed. The industry leaders who act now, by strengthening defences, piloting new identity models, and demanding industry-wide dialogue, will be best positioned to shape a future where innovation and trust can coexist.

 

About the author

Jeff Otto is Chief Marketing Officer at Riskified, a leader in AI-driven fraud prevention and risk intelligence for ecommerce. He brings deep expertise in marketing strategy, business development, and customer engagement, having held leadership roles at Marqeta, Salesforce, Morgan Stanley and Merrill Lynch. Passionate about emerging technologies, he frequently shares insights on ecommerce, customer experience, and fraud tactics through speaking engagements, webinars, and publications.

 

About Riskified

Riskified (NYSE:RSKD) empowers businesses to unleash ecommerce growth by outsmarting risk. Many of the world’s biggest brands and publicly traded companies selling   online rely on Riskified for guaranteed protection against chargebacks, to fight fraud and policy abuse at scale, and to improve customer retention. Developed and managed by   the largest team of ecommerce risk analysts, data scientists and researchers, Riskified’s AI-powered fraud and risk intelligence platform analyses the individual behind each   interaction to provide real-time decisions and robust identity-based insights. Learn more at riskified.com.