Mastercard urges controls to build trust in AI agents
US e-commerce spending involving AI agents could reach $1t by 2030.
Companies deploying artificial intelligence (AI) agents must define whom the agents represent and what actions they can take to earn customer trust, a Mastercard executive said.
“Governance is key,” Aditi Sawhney, senior vice president for Asia-Pacific security solutions at Mastercard Asia/Pacific Pte. Ltd., told Asian Banking & Finance. “But particularly on agents, it’s about understanding, firstly, who the agent is representing.”
Companies should establish an agent’s identity and apply what Sawhney called zero-trust authority before expanding its permissions.
“What data can it access? What actions can it take? Test all the cases that could happen. Only then can authority be expanded, and trust,” she said in a Microsoft Teams call.
An August report by McKinsey & Company, Inc. found that 40% of respondents from big organisations, defined as those with annual revenue above $1b, were scaling AI agents, up from 27% in 2025.
A separate McKinsey poll published in March found that only 32% of 496 organisations had adequate governance for agentic AI or were evaluating compliance and documenting processes.
Sawhney said AI development was outpacing governance at some companies. She said governance should be built into the development, testing, and deployment of AI systems.
Agentic AI and commerce also require companies to verify a buyer’s authority, not just protect customer credentials, she said.
Customers expect agents’ actions to be traceable, whilst companies need to be able to audit them when transactions fail, Sawhney said.
Companies also need payment providers, merchants, card issuers, and other parties to recognise authorised agents and distinguish them from rogue ones.
“You need all of the parties in the ecosystem to recognize when an agent is acting and to trust the information that flows with it so that they don’t block [the transaction],” she said.
“The ecosystem is going to need a lot of intelligence to be able to understand which is actually an agentic transaction and which is a rogue agent,” she added.
The use of AI agents could also generate revenue from transactions they initiate. US e-commerce spending involving AI agents could reach as much as $1t by 2030, McKinsey said in an October 2025 report.
Business-to-business transactions conducted through AI agents could reach $15t by 2030, Sawhney said, citing a Gartner, Inc. study published in October 2025.
She expects agents to move beyond single transactions based on individual instructions towards handling broader objectives within set controls and limits. Users may eventually allow agents to “drive an outcome,” she said, whilst retaining practical restrictions.