Canada's five biggest banks are already testing it, even as most clients stay wary
Five of Canada's biggest banks are now testing technology that lets an artificial intelligence agent reach into a cardholder's account and spend on their behalf, even though only 27 percent of Canadians know what agentic AI is.
BMO, CIBC, RBC, Scotiabank and TD have joined Visa's Agentic Ready program in Canada, which gives issuers a controlled environment to validate agent-initiated payments using live cards and real merchants, according to LeapRate.
Participants can test card enrolment, tokenization and transaction authorization before AI-driven commerce scales more broadly.
The pilot lands ahead of consumer demand.
A Visa Canada survey found 77 percent of Canadians believe AI will improve their lives, yet 92 percent expressed concern about privacy and ethical use.
A separate TD Bank survey of more than 2,500 consumers found AI use for personal finance jumped from 10 percent a year earlier to 55 percent in 2026, a shift the bank's Ted Paris called an "inflection point," Banking Dive reported.
That comfort has limits.
Trust drops sharply when consumers are asked to let AI make autonomous, high-stakes financial decisions, even as most accept it for fraud detection and spending tracking, according to the ABA Banking Journal's reporting on the TD survey.
Research from consulting firm Bain & Company, cited by CBC News, found about half of American respondents were not comfortable with AI completing transactions for them.
The infrastructure is arriving regardless.
Visa's Intelligent Commerce Connect lets merchants accept agent-initiated payments through a single integration, while Mastercard's Agent Pay registers and authenticates agents so each purchase carries the limits a user sets.
Stripe and OpenAI built the protocol behind ChatGPT's Instant Checkout, which uses a token so the agent never sees a buyer's card credentials.
PayPal and Microsoft Copilot Checkout have launched comparable services.
Retail platforms are already flipping the switch.
Amazon's Alexa for Shopping now lets US users buy a product automatically once its price drops below a set threshold, and Google's payments protocol will let AI tools purchase items when criteria such as brand and price are met, CBC News reported.
Google's Josh Woodward likened the guardrails to handing a teenager a first debit card with limits attached.
The brokerage side shows where this leads.
Last week, Wealth Professional reported the launch of Agentic Trading and an Agentic Credit Card by Robinhood, letting AI agents trade stocks and make purchases on a user's behalf.
The company warns that agents can make errors and that monitoring the account stays the user's responsibility.
Those errors are where the risk concentrates.
Cleber Martins of ACI Worldwide warns agentic commerce could fuel first-party fraud, because behaviour-based detection weakens when bots transact, BankInfoSecurity reported.
Experian's 2026 fraud forecast names agentic AI a leading threat and cites FTC data showing consumers lost more than US$12.5bn to fraud in 2024.
EY Canada counters that the same autonomy can power real-time fraud monitoring that cuts false positives.
Cybersecurity analyst Ritesh Kotak told CBC News the moves so far are "baby steps," but warned that recovering money after a bad agent purchase could prove difficult.
Interac's Shenela Tavarayan framed the core tension as balancing convenience against losing control of a bank account.
Scotiabank's Martin Ho drew the line at price, noting that letting AI buy $10 paper towels differs sharply from a $10,000 trip.
E-commerce analyst Jeremy Goldman offered a simpler caution to CBC News: many people shop for pleasure, and what they buy still says something about them.
