When tickets for your favourite artist go on sale in the middle of a critical meeting, the idea of dispatching an AI agent to buy them for you sounds pretty appealing. But how do you know it will buy the right tickets, stick to your budget, and keep your payment information safe?
“Trust is foundational to agentic commerce being able to truly scale.”
Answers to those questions are becoming more important as agentic commerce scales in Canada and AI agents move from recommending purchases to making them. A Mastercard report predicts that by 2030, more than one in 10 online shoppers will routinely use agents to shop and pay.
Making those purchases secure takes a lot of work behind the scenes, and trust built into every side of the transaction. As the buyer, you need to trust the agent to follow your instructions. The seller has to know it’s legitimately acting on your behalf, while the financial institution needs to verify that the purchase matches what you authorized before approving the payment. All of this must happen in real time, with safeguards against fraud and attempts to manipulate the agent.
Canada is already putting the technology to the test. This fall, Mastercard worked with AI company Flybits and credit card issuer Rogers Bank to complete its first Canadian transaction through Agent Pay, its agentic commerce service. The transaction went beyond showing that an agent could make a purchase. The participants were able to verify that the agent was legitimate, had permission to act, and that the transaction amount matched the consumer’s authorized intent, said Amisha Parikh, vice-president, security solutions at Mastercard, Canada, in an interview with Betakit. “I think the real learning here was that it works,” she said.
“Human-led, agent-executed” transactions
For Canadian startups building AI tools, selling online or developing payment technology, that first purchase by an AI agent offers an early look at what it takes to make these transactions secure.
Parikh described the arrangement as “human-led, but agent-executed,” with the customer deciding what to buy and how much to spend.
In her concert-ticket example, you could specify two seats for a Toronto show, the dates you are available, your preferred sections, and a $500 budget. The agent has room to find tickets that fit. It should not decide that a more expensive show in another city is close enough.
“The agent is executing, but the human has set those guardrails,” she said.
Businesses already use tools such as multifactor authentication and biometrics to check that you, as the buyer, are who you say you are. When an agent shops on your behalf, they also need to confirm that it genuinely represents you and has permission to make the purchase.
Parikh said by using Agent Pay, the agent used is legitimate and has the customer’s permission to make a purchase on their behalf. “Verifiable intent becomes very important,” she said.
What if someone interferes?
Even an authorized agent can still become a target for attack. Just as someone could take over a customer’s account, they could potentially try to redirect an agent by giving it different instructions—an attack known as a “prompt injection.”
The risks also extend to where it shops. A fraudulent merchant could appear to offer exactly what the agent has been asked to find and draw it into a scam.
As merchants introduce agents of their own, software could interact on both sides of a purchase. Parikh says it becomes important to establish which person or business each agent represents and whether it is following their instructions.
Existing fraud controls also need to evolve to distinguish legitimate AI agents from malicious bots and recognize when an agent is acting with a customer’s authorization, she said.
This is where Mastercard’s Agent Pay Trust Framework helps address the need to recognize legitimate agents and assess the risks around their purchases. That means understanding who an agent represents, what it has permission to do, and whether the merchant can be trusted.
Mastercard brings together insights about identity, customer intent, behavioural patterns, and fraud risk to help banks and merchants better understand agent-led purchases. Parikh said the challenge is turning those signals into timely, actionable intelligence: helping to distinguish legitimate activity from potential threats, identify when additional checks are needed, and reduce unnecessary friction for customers.
Canada and the future of commerce
Parikh said advancing agentic commerce in Canada will require cooperation among financial institutions, retailers, government, academia, and payment networks.
She pointed to Mastercard’s Global Intelligence and Cyber Centre of Excellence in Vancouver as an example of how talent, research and partnerships contribute to advancing work on digital trust and intelligence.
“As AI capabilities evolve, she said, the safeguards around them must evolve, too. For agentic commerce, that means maintaining confidence that agents are acting with permission, within agreed limits, and with clear accountability.”
“Trust is foundational to agentic commerce being able to truly scale,” she said.
PRESENTED BY
Learn more about Mastercard’s approach to cybersecurity, digital trust and the future of commerce.
Feature image courtesy Mastercard