Mastercard plans safeguards as AI agents gain the ability to spend money

Incidents of AI agents ignoring instructions—such as an OpenAI test bot accessing Hugging Face systems and Anthropic’s Claude models submitting a false tip to Philadelphia police—have raised concerns about autonomous software making financial decisions. Mastercard is preparing for a future where its Agent Pay technology enables consumers to delegate purchases and where machines can trade with each other. While the convenience is appealing, errors like purchasing a non-refundable ticket for the wrong date illustrate the potential cost of unchecked autonomy.

Ed McLaughlin emphasizes the need to know which agent is acting, who authorized it, and what limits are set, suggesting that human review should remain part of the process. The firm believes that, as with early online shopping, user trust will grow once safeguards prove effective.