The Identity Crisis: Why Agentic Commerce Can't Scale Without a Trust Layer
AI-generated image · Bay Street Wire
As AI agents move from recommending products to executing checkouts, the industry faces a critical hurdle: verifying that an agent is legitimate and authorized to spend a consumer's money.
From an operator's perspective, the promise of agentic commerce—where AI handles the search for a product and executes the purchase—is a powerful lever for conversion. However, as a tech columnist covering the retail beat, I have to be clear: as BetaKit recently reported, this is a non-starter for retailers until we solve the identity and authorization layer. Without a rigorous framework to verify who an agent represents, we are inviting massive fraud into the heart of the transaction.
***
**Opinion: The Authorization Gap**
While the technical ability for an AI to click "buy" exists, the commercial viability does not. If we cannot distinguish between a legitimate agent acting on a customer's behalf and a malicious bot designed to exploit a system, the risk outweighs the reward. The industry needs a standardized, verifiable identity layer before this can move from a pilot project to a retail standard.
***
**The Scale of the Shift**
The shift toward autonomous shopping is already accelerating. A Mastercard report predicts that by 2030, more than 10% of online shoppers will routinely utilize agents to handle both the shopping and the payment process. This marks a fundamental change in the consumer journey: AI agents are moving beyond simple recommendations to making actual purchases.
**The Trust Framework: Human-Led, Agent-Executed**
BetaKit highlights that for agentic commerce to scale, trust must be embedded into the transaction. This requires a three-way verification: the buyer must trust the agent's adherence to budget, the seller must verify the agent's legitimacy, and the financial institution must confirm the purchase aligns with authorized intent.
Amisha Parikh, vice-president of security solutions at Mastercard, Canada, describes this model as "human-led, but agent-executed." In this scenario, the human sets the guardrails—such as a $500 budget for concert tickets—and the agent executes within those limits. Parikh told BetaKit that "verifiable intent" is a critical component.
**Real-World Testing in Canada**
This theory is already being tested. BetaKit reports that this fall, Mastercard collaborated with credit card issuer Rogers Bank and AI company Flybits to execute the first Canadian transaction using Agent Pay, Mastercard's agentic commerce service. Parikh noted that the trial proved a transaction could be completed while verifying the agent's legitimacy and permission, concluding simply that "it works."
**The New Frontier of Fraud**
However, agentic commerce introduces novel vulnerabilities. BetaKit outlines risks such as "prompt injection," where malicious actors redirect agents with fraudulent instructions, and "fraudulent merchants" designed to lure agents into scams.
To combat this, Parikh suggests fraud controls must evolve to distinguish between malicious bots and authorized agents. Mastercard is addressing this through its Agent Pay Trust Framework, which aggregates data on identity, behavioral patterns, and intent to provide banks and merchants with actionable intelligence.
**The Path Forward**
Scaling this technology will require a coordinated effort between government, academia, retailers, and financial institutions. Parikh pointed to Mastercard’s Global Intelligence and Cyber Centre of Excellence in Vancouver as a hub for the research necessary to build this digital trust. As AI capabilities grow, safeguards must evolve in tandem to ensure clear accountability and strict adherence to consumer limits.