Agentic Commerce Is Coming, but Zac Cohen Says the Trust Layer Is Not Ready
Agentic commerce has started to move into live use, and the payment industry is beginning to confront a problem it has not had to solve at this scale before.
When an AI agent acts for a customer, the transaction is no longer only about verifying the person behind the account. The system also has to understand the authority given to the agent carrying out the task.
Zac Cohen, Chief Product Officer at Trulioo, believes that gap is where the industry needs to focus.
Zac Cohen
“I think the industry is ready, but the trust framework isn’t,” Zac said in an interview with Fintech News Network.
AI agents in commerce are beginning to influence the work that happens before money moves.
They can interpret customer preferences, narrow choices and bring a transaction closer to completion with less direct involvement from the buyer.
As that role grows, financial institutions need a clearer way to recognise whether the agent was properly authorised and whether its actions still match the limits set by a real person or business.
“It’s not a feature issue that we’re solving for anymore,” Zac added. “It’s really the governance and trust framework.”
He believes today’s volumes remain small, but adoption could accelerate once consumers become more comfortable allowing agents to act on their behalf.
Many financial institutions still treat the space as early, even as the operating models needed to govern it are only beginning to take shape.
Agentic Commerce Could Reshape the Payment Relationship
Agentic commerce brings the payment decision closer to the search and selection process.
A customer may still give the mandate, but an AI agent can begin carrying that mandate through the journey before a payment instruction reaches the bank or network.
The more this happens, the less useful it becomes to look only at the final payment event.
Payment firms or even a bank that only sees the transaction at the point of payment may miss how the agent shaped the decision earlier.
If the agent has already interpreted the customer’s instruction and moved the purchase towards completion, trust has to be established earlier in the journey.
Zac’s warning is that companies which move too slowly could lose more than a new revenue stream.
“The laggards aren’t only going to lose potential revenue opportunities, but they might be disintermediated completely,” he stated.
Treating agentic AI as a front-end upgrade would miss much of the challenge.
A better interface does not tell an institution whether the agent had the authority to act, whether the instruction came from a real customer, or whether the action stayed within the intended limits.
Zac Cohen said that issuers should be part of this conversation early because banks already hold the customer relationship around money movement, identity and account access.
As agents begin to act on behalf of consumers, that position could help issuers shape the trust framework rather than simply process the payment after another layer has made the decision.
Know Your Agent Enters the Conversation
Financial institutions already have established ways to verify the people and businesses that use their services.
Know Your Customer (KYC) helps confirm whether an individual is who they claim to be, while Know Your Business (KYB) helps institutions verify the business behind an account or transaction.
Agentic commerce changes that chain by placing an AI agent between the person giving the instruction and the transaction being carried out.
Zac sees that shift creating the need for a new layer of verification around the agent itself.
The industry has begun referring to that layer as Know Your Agent, or KYA.
The idea remains early, but the logic follows from the systems that financial institutions already use.
If KYC helps establish the person and KYB helps establish the business, KYA asks whether the agent carrying out the instruction had the authority to act in the first place.
“KYC is for consumers, KYB is for businesses, and KYA is for non-human actors,” he mentioned.
He also highlighted that the shift changes the question institutions need to answer when an agent is involved in a transaction.
“We’re no longer asking who is behind this transaction,” he told. “We’re asking what, and what had the right to command, what had the right behaviour, what had the right decision.”
Institutions also need different evidence.
Traditional checks rely on documents and records that tie a person or business to a known identity, while an AI agent has to be assessed through its origin, instructions and operating limits.
The Chief Product Officer at Trulioo said a KYA framework should show who directs the agent and what boundaries govern its actions.
Institutions still need to understand the person or business behind the agent, but they also need to examine the code and instructions shaping its behaviour.
Agentic payments could push institutions to verify more than the customer at onboarding, requiring them to monitor whether the agent stays within its original authority.
Agentic Payments Bring a Different Fraud Problem
The need to verify an agent’s authority also changes how institutions think about fraud.
Financial crime teams already watch for impersonation, account takeover, mule accounts and suspicious transaction behaviour, but agentic payments create another point of exposure inside the decision-making process itself.
Zac also mentioned that the attack surface changes when an AI agent becomes the actor carrying out the instruction.
A bad actor may not need to pretend to be the customer if they can influence what the agent believes it should do.
“We’re no longer worried about a bad actor impersonating someone else,” said Zac. “We’re actually worried about someone changing what an agent thinks is the right thing to do.”
@fintechnewsnetwork
The threat is no longer someone pretending to be you. It’s someone changing what an AI agent thinks it’s supposed to do. Zac Cohen, Chief Product Officer at Trulioo, explains why injection attacks and machine-speed exploitation are the fraud risks the industry needs to prepare for now. #fintech #AIfraud #agenticAI #cybersecurity #FintechNews
♬ original sound – Fintech News Network – Fintech News Network
That could involve changing the instruction, altering the agent’s guardrails or steering the agent towards an action the customer did not intend.
The concern grows when agents operate quickly across large transaction volumes, giving attackers room to exploit one weak point before traditional review processes catch up.
Zac suggested that fraud and financial crime teams will need stronger intelligence signals inside each transaction.
Procedures built for human-led payment journeys may struggle when an autonomous or semi-autonomous agent can interpret instructions and act within seconds.
The same AI tools that make agentic commerce possible may also need to support the teams managing risk.
Fraud prevention and compliance teams will need systems that can detect unusual agent behaviour early, before abuse scales through automated transactions.
Regulators Are Sketching the Boundaries
Regulators are beginning to look more closely at agentic AI, although Zac sees a gap between policy ambition and what the market can implement today.
He pointed to Singapore’s work on agentic AI governance, which gives the industry an early view of how firms can manage autonomy with clearer oversight and explanation.
The direction is helpful, but turning policy language into working controls will take time, especially while agentic systems are still developing.
Zac said regulators are mainly trying to make sure consumers can use new technology safely and that financial institutions know how to respond when something goes wrong.
Agentic AI makes that harder because the systems being governed are still taking shape, even as commercial use cases move ahead.
Financial institutions will need to build automation in a way that customers can use easily and that regulators can still examine with confidence.
Firms that treat innovation and governance as separate tracks may find it harder to scale agentic commerce safely.
Governance Has to Come Before the Feature Race
“Don’t start with focusing on the capability,” Zac added. “Start with the governance framework.”
Agentic commerce may look simple to the customer, but banks and payment companies need to prove that the agent had the right authority to act.
They need to build trust into the system before agentic payments scale, especially if they want to stay close to the customer relationship rather than leave that role to another layer.
Zac Cohen shares more on agentic commerce, Know Your Agent and the future of trusted AI-led payments in the full interview here.
The post Agentic Commerce Is Coming, but Zac Cohen Says the Trust Layer Is Not Ready appeared first on Fintech Singapore.
Read More