Companies adopting artificial intelligence (AI) agents must establish clear guidelines defining whom these agents represent and the extent of their authority to take action, as these measures are essential to building customer confidence, according to a Mastercard executive.
“Governance is key,” said Aditi Sawhney, senior vice president for Asia-Pacific security solutions at Mastercard Asia/Pacific Pte. Ltd. She highlighted the importance of identifying the entity an AI agent represents before allowing it to operate.
Businesses should first establish an agent’s identity and introduce a zero-trust approach to its permissions before granting broader access.
This involves determining the information an agent can access, the activities it is authorised to perform and testing different possible scenarios. Only after these checks should companies consider expanding its authority and level of trust, Sawhney said during a Microsoft Teams interview.
An August report by McKinsey & Company found that 40% of respondents from large organisations with annual revenues exceeding $1b were scaling their use of AI agents, compared with 27% in 2025.
However, a separate McKinsey survey published in March found that only 32% of 496 organisations had established sufficient governance frameworks for agentic AI or were assessing compliance requirements and documenting related procedures.
Sawhney noted that AI adoption was advancing faster than governance frameworks in some businesses. She stressed that governance measures should be integrated throughout the development, testing and deployment stages of AI systems.
As agentic AI becomes more widely used in commerce, companies must also verify whether an AI agent has the authority to make purchases or conduct transactions, rather than focusing solely on protecting customer credentials.
Customers need visibility into the actions performed by their AI agents, while businesses must be able to trace and audit those activities when transactions encounter problems, she said.
This will require payment providers, merchants, card issuers and other participants across the ecosystem to identify legitimate AI agents and distinguish them from unauthorised or malicious ones.
Sawhney emphasised that all parties involved must be able to recognise agent-driven transactions and trust the information accompanying them to prevent legitimate payments from being declined. She added that the wider ecosystem would need advanced intelligence to differentiate genuine agentic transactions from those initiated by rogue agents.
The growing adoption of AI agents could also create new commercial opportunities through purchases and transactions initiated on behalf of customers. According to a McKinsey report published in October 2025, US e-commerce spending involving AI agents could reach $1t by 2030.
Meanwhile, business-to-business transactions conducted through AI agents could total $15t by the same year, Sawhney said, citing a Gartner study published in October 2025.
She expects AI agents to gradually evolve beyond executing individual transactions based on specific instructions to managing broader objectives within predefined boundaries and permissions.
In the future, users may allow AI agents to independently pursue particular outcomes, while retaining practical safeguards and restrictions over the decisions and actions they can undertake.
Click here for more on Banking













