The Identity Gap: Why Agentic Commerce Fails Without Verifiable Intent
AI-generated image · Bay Street Wire
Mastercard's recent Canadian pilot proves that while AI agents can execute a checkout, the real battle is solving the authorization and trust framework required for merchants to accept them.
### The Promise of the Autonomous Checkout
Imagine dispatching an AI agent to secure high-demand concert tickets while you are trapped in a business meeting. This is the core of agentic commerce—a shift from AI that merely recommends products to AI that actively executes transactions.
As BetaKit first reported, the scale of this shift is projected to be significant. A report from Mastercard predicts that by 2030, more than 10% of online shoppers will routinely rely on agents to both shop and pay. However, the industry must first solve a fundamental operational hurdle: how a merchant knows an agent is legitimately acting on a human's behalf.
### The 'Human-Led, Agent-Executed' Model
To test this, Mastercard recently collaborated with AI firm Flybits and credit card issuer Rogers Bank to facilitate the first Canadian transaction via Agent Pay, Mastercard's agentic commerce service.
Amisha Parikh, vice-president of security solutions at Mastercard, Canada, described this model to BetaKit as "human-led, but agent-executed." In this framework, the human remains the decision-maker. For example, a user might specify a $500 budget for two seats at a Toronto show; the agent is tasked with finding tickets within those exact parameters, without the autonomy to choose a more expensive show in a different city.
The pilot proved that the technology works, but the primary achievement was the verification process. Participants confirmed the agent was legitimate, possessed the necessary permission, and that the transaction amount aligned with the consumer's authorized intent.
### Solving the Authorization Crisis
Current security stacks rely on biometrics and multifactor authentication for humans. When a bot handles the checkout, these signals vanish. Parikh told BetaKit that "verifiable intent" is the critical missing link. For a transaction to be secure, three parties must reach a real-time consensus: the buyer must trust the agent, the seller must verify the agent's legitimacy, and the financial institution must confirm the purchase matches the authorized intent.
### New Risks and the Path to Scale
Security risks also evolve. BetaKit reports that authorized agents can be targeted by "prompt injection," where third parties attempt to redirect an agent with conflicting instructions. Additionally, fraudulent sellers could create storefronts designed specifically to lure AI agents into scams.
To combat this, Mastercard is deploying its Agent Pay Trust Framework, which integrates data on identity, behavioral patterns, customer intent, and fraud risk to provide banks and merchants with actionable intelligence.
Scaling agentic commerce in Canada will require deep cooperation across payment networks, retailers, government, academia, and financial institutions. Mastercard is leveraging its Global Intelligence and Cyber Centre of Excellence in Vancouver to further research this digital trust. As Parikh noted, trust is the foundation upon which agentic commerce must be built.