Agentic ecommerce brings new opportunities for merchants, but also new questions about trust, intent, and authorisation. Based on research by The Paypers and Riskified, here are five risks payments and fraud teams should prepare for.
The shift from search-driven shopping to agentic ecommerce is already creating exciting opportunities for digital merchants. Walmart’s Sparky assistant within ChatGPT is driving unexpected new purchase journeys. Shopify’s embrace of agentic is expanding LLM access to a wide range of merchants. And a recent survey shows that nearly half of consumers have already made a purchase influenced by AI.
However, agentic ecommerce also raises new, fundamental questions about trust, intent, authorisation, fraud, and risk.
That’s why we collaborated with risk intelligence experts at Riskified on an exclusive new research report exploring the benefits and risks of agentic ecommerce for digital retailers.
The report examines emerging agentic models, consumer attitudes, and the technologies shaping this new ecosystem. It also carefully examines the risk factors that payments and fraud teams will need to consider from the outset as they make vital technology and policy decisions.
For a preview, here are five risks to watch according to The Paypers research with Riskified:
Signal loss will disrupt fraud detection
When an AI agent does the shopping, many approval decisioning signals disappear or become unreliable. Indicators like mouse movements, device information, and IP location become less meaningful.
Riskified research indicates that up to 34% of the data used by existing AI models for fraud decisions was missing from ChatGPT's Instant Checkout. What data will replace what gets lost?
Third-party agents introduce new risks
Third-party AI agents can bring merchants new customers, but they also mean giving up some control over the shopping journey, customer relationship, and risk signals. Store-native agents have their own downsides, but they do offer greater control.
Merchants must be proactive and weigh reach, control, and risk factors. ‘Wait and see’ will be costly.
New fraud attack surfaces will emerge
While signal loss and AI-enabled schemes complicate detection of traditional fraud, payments teams will also have to contend with novel attack vectors, such as:
- Compromised or hostile agents;
- Agent takeover;
- Stolen payment credentials being exploited through agents;
- Prompt injection through product listings, reviews, or other web content;
- Agents being manipulated into purchasing the wrong product, excessive quantities, or from a particular seller.
Verifiable intent remains elusive
How does a merchant know that an agent is actually following the consumer's instructions? What constitutes ‘proof of intent’ in a payments context? Initiatives like Mastercard's Verifiable Intent, Google's AP2, Visa's Trusted Agent Protocol, and IXOPAY/ZIP's Unified Trust Layer attempt to establish verifiable links between consumer intent, agent actions, and payment authorisation. But the landscape remains fragmented and evolving. For merchants, the question is which standards and trust mechanisms will ultimately prevail.
Chargebacks will get more complicated
If an agent misunderstands an instruction, hallucinates, buys the wrong product or is manipulated, who’s responsible? How will merchants handle agent-involved transaction disputes?
Agentic ecommerce may shift the liability calculus when customers deny having authorised purchases made by agents, making it even more challenging to balance chargeback costs with customer experience.
Explore the research
For strategies to capture the moment, manage risk, and more, download the full report: The benefits & risks of agentic AI in ecommerce.
About the author
Vlad is a Senior Writer at The Paypers, working in the commercial team. He uses his research, content, and people skills for all editorial activities revolving around payments, banking, and fintech. Vlad has a degree in Biology and Molecular Genetics and an extensive background in creative writing. You can reach out to him on LinkedIn.
