Mastercard executive tells firms to set limits for AI agents
In a March poll, only 32% of 496 firms had sufficient governance for AI agents.
Aditi Sawhney, a senior vice president at Mastercard, told Asian Banking & Finance that firms must show who an AI agent works for. Firms must also show the steps that it can do. Sawhney tells that this gives customers cause to have trust. McKinsey tells that US spending through agents can be $1t at most by 2030.
What she proposes
Firms must first set the identity of an agent. They must give it small authority and test all conditions that can occur. More authority can come only after that.
Banks, merchants and payment providers must know which agents have authority and which are rogue. Customers want to follow each step. Firms must be able to audit a payment that does not work.
The numbers
A McKinsey report in August found that 40% of large firms used more agents, up from 27% in 2025. A McKinsey poll in March found that 32% of 496 firms had sufficient governance. Governance is the set of rules and checks for how a system is made.
What is not known
Sawhney gave no test results for the method. Sawhney tells that customers can then let agents work to a result. This is a view, not a measure.
Insight: The firm that sets agent trust rules also sells the tools
Sawhney tells that all parties must know when an agent works. Mastercard has made one tool for each part of this trust, and its chief executive tells that many merchants are not prepared. This reads as a move to become the check that all parties must use, and the first check in wide use can set the rules. Leaders must tell their provider which check it uses.
Sources
- Mastercard urges controls to build trust in AI agentsAsian Banking & Finance
Posted