Riskified's latest report captures a pivotal inflexion point in ecommerce: the shift from search-driven shopping to AI-mediated, and increasingly autonomous, purchasing. The findings paint a picture of an industry racing to capture enormous upside while operating without the fraud infrastructure or consumer trust needed to support it safely.
Two models, one emerging market
The report frames agentic commerce around two distinct architectures:
Store-native agents, such as Amazon's Rufus, Walmart's Sparky, and Macy's Ask Macy's, are merchant-owned assistants that guide discovery and post-sale support but currently stop short of completing transactions autonomously; the customer still approves the final purchase.
AI assistant agents, by contrast, operate across merchants, comparing options and potentially executing purchases directly on a consumer's behalf. This second category represents the more disruptive long-term shift, pointing toward a future dominated by agent-to-agent commerce rather than human browsing sessions.
This bifurcation forces merchants to design for two fundamentally different audiences simultaneously: humans seeking an immersive experience, and machines executing narrow, functional tasks.
The scale of the opportunity
McKinsey projects agentic commerce could orchestrate up to USD 1 trillion in US retail revenue by 2030, scaling to USD 5 trillion globally. AI-referred traffic, while still starting from a small base, grew exponentially year-on-year from 2024 to 2025: over 2,000% in fashion, nearly 1,200% in luxury, and almost 7,500% in speciality retail, according to BCG. Riskified's own research found 48% of consumers have already made a purchase influenced by AI search results.
Early movers are already seeing tangible payoff: Amazon credits Rufus with a USD 10 billion boost to annual sales, while Macy's reports that Ask Macy's users spend 4.75 times more than non-users. A live poll at Riskified's 2026 Ascend conference found 32% of merchants already have a native AI shopping assistant live or in development, with another 35% exploring options, signalling that a wait-and-see posture is rapidly becoming a competitive liability.
A trust deficit at the center of the model
Despite this momentum, only 16% of global consumers say they are comfortable with an agent spending their money autonomously, even with permission; 55% say they are not comfortable at all. Over half (54%) believe AI increases fraud risk, and when asked about their biggest concern with agent-led purchasing, 31% cited payment information security and 12% cited the risk of an agent being manipulated by a scammer into making fraudulent purchases.
This is where the report's central tension lies: agentic commerce requires removing the human from the transaction loop, but doing so eliminates the natural checkpoint that most fraud controls are built around. Without a customer actively confirming each purchase, merchants and payment systems must instead verify that an agent is acting on behalf of an authorised human, and that this authority was properly delegated.
New fraud vectors, new defences
The report identifies a new threat category rather than a simple extension of existing fraud. Agentic commerce disrupts established anti-fraud detection methods while simultaneously introducing novel vulnerabilities: hostile agent takeovers, prompt injection steering, credential theft via tokenised checkout flows, and fraudulent chargeback exploitation. The scale of criminal interest is already visible: underground forum mentions of AI agents grew 450% over six months in 2025.
Riskified's own testing of OpenAI's Instant Checkout in ChatGPT is a striking case study: the company's R&D team found orders being approved that a competent fraud stack should have declined, tracing the gap to the fact that 34% of the data typically used by machine learning fraud models was missing from these transactions. Inventory management and real-time pricing infrastructure, the report notes, also weren't yet ready for large-scale agentic usage.
Industry response: a race to standardise trust
In response, payments and technology providers are building competing frameworks to authenticate agents and verify intent, including Google's AP2 and Verifiable Intent mandates, OpenAI's ACP, Visa's Trusted Agent Protocol (TAP), and Mastercard's Agent Pay tokenisation. Alongside these protocols, third-party specialists like Riskified are developing dedicated ‘trust agents’ designed to operate natively within agent-to-agent workflows, combining identity intelligence, behavioural history, and transaction anomaly detection.
Bottom line
Therefore, agentic commerce is no longer speculative, and the revenue upside for early adopters is already measurable. But that upside is inseparable from a fraud and trust problem that current infrastructure has not solved. Riskified's recommendation is to treat fraud prevention as a first-principles design requirement, embedded from day one of any agentic deployment, rather than a feature bolted on after the fact.
Want to learn more? Download Riskified’s report: The benefits & risks of agentic AI in ecommerce
About Claudia Pincovski
Claudia is a News Lead Editor at The Paypers. Holding a bachelor’s degree in journalism, she is very passionate about exploring the latest news on financial inclusion, financial literacy, digital banking, and Open Finance. Claudia is a diligent researcher, a meticulous editor, and an active advocate for diversity and inclusion.
About Riskified
Riskified 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.
