SEON survey: 65% blame AI companies for agent payment fraud

Nearly two-thirds of US consumers who expressed a view on responsibility for fraudulent transactions carried out by AI agents said the company behind the technology should be accountable, according to new research from fraud prevention provider SEON.

The survey of 2,026 US adults found that 68% use an AI agent at least weekly for personal tasks, but willingness to let the technology handle money and sensitive financial information remains considerably lower.

Half of respondents said they were uncomfortable allowing an AI agent to use their credit card details to complete a purchase they had requested. Only 15% were willing to share financial information with an agent, while 32% said they would not share any personal information.

The findings highlight a potential obstacle for banks, payment providers and fintech companies developing systems that allow AI agents to initiate transactions or access customer accounts.

Responsibility for fraudulent transactions emerged as a particular concern.

Among 885 respondents who expressed an opinion, 65% said the company behind an AI agent should take responsibility if the agent completed a fraudulent transaction. Another 14% assigned responsibility to themselves, 11% to the merchant and 10% to their bank or card provider.

These figures reflect consumer expectations rather than established legal liability.

The research also identified uncertainty over whether consumers could recognise fraudulent instructions. Some 43% lacked confidence in their ability to determine whether an agent’s request for information or account access was legitimate, while 39% did not trust agents to identify suspicious situations on their behalf.

The confidence gap was particularly pronounced among older consumers. Some 61% of Baby Boomers reported little or no confidence in assessing the legitimacy of agent requests, compared with 28% of Millennials.

For financial institutions, the findings raise questions about authentication, transaction authorisation and the ability to distinguish legitimate automated activity from account compromise or fraudulent instructions.

SEON argues that identifying an AI agent is insufficient on its own. Financial institutions also need to assess device, identity and behavioural signals to determine whether activity is consistent with customer permissions.

Human support was the most frequently cited measure for increasing trust, selected by 50% of respondents, followed by reversible transactions at 42% and stronger identity verification at 39%.

The survey was conducted between 2 and 6 October 2026 and weighted to the US adult population.

It measures attitudes rather than actual fraud losses or transaction volumes. Nevertheless, it suggests that broader adoption of AI-driven payments may depend as much on demonstrable accountability and fraud safeguards as on the technology’s ability to complete transactions.