Updated October 13, 2026 4:24 pm
In short
American Express has added its intent-verification layer to the broader ACE developer toolkit, is already talking with payments partners about implementation, and says the framework is meant to work across different agentic commerce models as the category evolves.
- American Express introduced the Intent Intelligence Framework for AI agent purchases.
- The system captures customer intent before checkout and uses that record to validate payments.
- Amex says the framework could reduce unauthorized purchases, fraud, and dispute headaches.
- The company is folding the new tool into its ACE Developer Toolkit and seeking ecosystem partners.
- Consumer interest in AI shopping is rising, pushing payments firms to build guardrails now.
Update — October 13, 2026 4:24 pm
American Express says the new intent framework is part of its broader Agentic Commerce Experiences toolkit, which it launched in April. The company is already working with partners across the payments ecosystem to implement those tools.
Amex also said it wants the framework to be flexible enough to work across different AI models and agent-led checkout flows, not just a single type of assistant. It is presenting the system as a possible industry standard as AI-driven purchases become more common.
Bozian also said Amex plans to keep developing the ACE platform as agentic commerce evolves, including features that could help track which agent authorized a purchase and when.
American Express on Tuesday introduced a new payment framework designed to verify a customer’s purchase intent before an AI agent can complete a transaction, a move aimed at reducing unauthorized or mistaken purchases as agentic commerce expands. The system, called the Intent Intelligence Framework, is meant to help ensure that what a consumer asks an AI assistant to buy is exactly what gets charged to their card.
The launch matters because AI shopping agents are moving from novelty to real payment activity, but the large language models behind them can still misunderstand instructions, improvise, or drift away from what users authorized. Amex’s answer is to move the approval and verification step earlier in the checkout flow, before a transaction becomes a dispute.
What American Express is trying to solve
American Express says the challenge is not hypothetical: as AI agents take on more shopping and booking tasks, they will increasingly sit between the consumer and the merchant. That creates a new transaction model with three distinct parties instead of two — the customer, the agent, and the seller — and each of those parties needs to be verified in a way that traditional card payments were never designed to handle.
The company’s concern is that agentic systems can be nondeterministic. In plain terms, that means they do not always behave in a perfectly predictable way. An agent might interpret a prompt loosely, pick the wrong item, choose the wrong retailer, or carry out a purchase that fits part of the instruction but not the full set of requirements.
For a payment network, that creates a familiar but more complicated problem: fraud prevention and dispute handling become harder when the human never directly sees the final checkout step. Amex is trying to build an infrastructure layer that preserves the user’s original intent all the way through payment authorization.
How the Intent Intelligence Framework works
The framework starts by capturing purchase intent before a payment is approved. American Express says the process begins with a Create Intent API, which records the customer’s buying instructions and any constraints they specify, such as price ceilings, merchant preferences, product type, size, or brand requirements.
From there, the customer authenticates themselves and signs off on the intent that the agent collected. That approval helps establish a clear record of what the user actually authorized, rather than relying on a post-purchase explanation if something goes wrong.
Once intent is confirmed, the system creates two key records: an Intent ID and a Proof of Intent Token. The Intent ID acts as a unique reference for the transaction’s approved purpose, while the token carries the high-level rules attached to that permission. Amex says that token is then used to obtain a secure digital payment credential for checkout instead of exposing the customer’s actual card number.
In theory, that means the payment can proceed only if the agent stays within the boundaries the customer set. If the agent strays, the transaction can be blocked before it settles.
| Framework step | What happens | Why it matters |
|---|---|---|
| Create Intent API | Captures the customer’s purchase instructions and constraints | Defines what the user meant to authorize |
| User authentication | Customer verifies identity and approves the captured intent | Links the request to a real, consenting card member |
| Intent ID | Creates a unique identifier for the approved purchase | Provides a trackable record for the transaction lifecycle |
| Proof of Intent Token | Stores key buying constraints and produces a secure payment credential | Lets the agent pay without using the card number directly |
| Checkout validation | Checks the payment against the original intent | Helps stop unauthorized or mismatched purchases |
Why American Express is moving early
Amex says it has not yet seen a noticeable rise in disputes tied to agentic commerce, but it believes the time to build safeguards is before the problem becomes widespread. That is a common strategy in payments: once a new spending behavior scales, the cost of retrofitting controls can be much higher than designing them upfront.
The company is also reacting to consumer interest. In an Amex Trendex survey cited by the company, 66% of U.S. consumers said they are likely to use AI agents to search for the best price on a product or service over the next 12 months, while 61% said they are likely to use AI for product or service recommendations. Those numbers suggest that shoppers are increasingly comfortable letting software influence or even carry out purchasing decisions.
That consumer openness helps explain why payment companies, card issuers, and merchants are racing to define the rules. If AI agents become routine intermediaries, the financial infrastructure has to know whether a bot is merely assisting a shopper or actually spending on the shopper’s behalf.
What makes this different from ordinary card authorization?
It is different because the framework tries to validate the meaning of the purchase, not just the payment credentials. Traditional card authentication focuses on whether the person or device making the transaction is permitted to use the card. Amex’s system goes further by asking whether the transaction still matches the user’s original instructions.
That distinction matters most in edge cases. If a user tells an agent to order white Nike sneakers in size 9 and the agent instead tries to buy Adidas shoes, the card network could have enough context to reject the transaction outright. The same could apply if the order exceeds a stated budget or comes from a merchant the customer did not approve.
American Express says its goal is to keep agent-driven purchases “seamless, trusted, and secure,” while making sure the customer’s instructions survive from the first request to the final payment.
How the company plans to use the data trail
The Intent ID and token are not only meant to approve or deny a payment. Amex says they can also support later stages of the transaction process, including fraud prevention and dispute resolution. That could give issuers and merchants a richer paper trail when a customer claims that an AI assistant bought the wrong item or exceeded permission.
In a more autonomous future, that record could become especially important. Bozian, Amex’s vice president of agentic partnerships and strategy, said the company expects some future purchase flows to involve less direct user oversight. For recurring, low-stakes items such as paper towels or toilet paper, a consumer may care more about size and price than the merchant name. In those cases, automatic purchasing can make sense — but only if the system can verify that the purchase still fits the customer’s boundaries.
That is where the framework’s audit trail becomes valuable. A log showing which agent received permission, what it was allowed to do, and when it acted could help determine whether a charge was legitimate, mistaken, or fraudulent.
Why agentic commerce is becoming a payment industry issue
AI shopping tools are no longer limited to recommendation widgets or chat-based product search. They are increasingly being built to take action: compare products, apply filters, choose merchants, and in some cases finish the transaction. Once those tools can complete purchases, payment companies have to decide how to handle authorization, liability, and fraud in a world where the consumer is not the only actor involved.
That is why Amex is not treating this as a narrow product feature. The company wants the framework to become a standard for AI-powered transactions across the broader ecosystem. Standards matter in payments because merchants, processors, wallet providers, and card networks all need a common language to interpret what a user approved.
Without that shared infrastructure, the risk is fragmentation: one AI shopping system might collect intent one way, another might do it differently, and merchants would be left trying to decipher inconsistent signals. Amex’s pitch is that an interoperable framework could make agent-driven commerce safer and easier to scale.
Where Amex’s broader agentic strategy fits in
The new framework builds on the company’s earlier work in agentic commerce. In April, American Express launched its Agentic Commerce Experiences, or ACE Developer Toolkit, which includes services that help partners build agent-powered shopping flows. The Intent Intelligence Framework is now being folded into that toolkit.
Amex says it is already in conversations with partners across the payments ecosystem to deploy ACE and its agentic tokens, which contain the company’s payment credentials. Those discussions suggest the company wants this to be more than a closed internal experiment. Instead, it appears to be laying groundwork for a wider ecosystem of tools that merchants and developers can adopt.
The company also says ACE will continue evolving alongside the market. That is important because agentic commerce is still at an early stage, and the payment requirements may change as AI models, consumer habits, and merchant integrations mature.
What this means for shoppers and merchants
For consumers, the biggest promise is convenience without surrendering control. If AI agents can search, compare, and purchase on a user’s behalf, they could save time and potentially find better prices. But that only works if customers are confident the agent will not make an unauthorized or off-spec purchase.
For merchants, the upside is a potentially smoother transaction flow with less friction at checkout. The downside is that any error in the intent-capture process could lead to blocked sales or additional verification steps. Merchants will also need to think carefully about how agent-generated purchases appear in their systems, especially if the buyer is acting under constraints rather than making a fully manual decision.
For payment companies, the stakes are broader still. The emergence of agentic commerce could force the industry to rethink the mechanics of authorization, authentication, and disputes. That makes frameworks like Amex’s less about one company’s product roadmap and more about the plumbing of the next generation of digital commerce.
What comes next
American Express plans to keep developing the ACE ecosystem as agentic commerce grows, and the company is set to discuss the new framework in more detail during TechCrunch Disrupt 2026. At that event, Bozian is expected to explain how the system works and why the company believes it can support the future of AI-enabled shopping.
The larger question is whether other payment networks and commerce platforms will adopt similar models. If they do, intent verification could become a standard layer of AI transactions, much like two-factor authentication became routine in account security. If they do not, the industry may fragment into multiple incompatible ways of defining what an AI agent is allowed to buy.
Either way, Amex’s move signals that the payment industry is no longer waiting for agentic commerce to become a problem. It is building guardrails now, before the technology has a chance to spend first and ask questions later.
Key facts at a glance
| Item | Details |
|---|---|
| Company | American Express |
| Announcement | Intent Intelligence Framework for AI agent purchases |
| Purpose | Verify purchase intent before payment is completed |
| Launch context | Expands Amex’s ACE Developer Toolkit from April |
| Consumer survey cited | 66% likely to use AI for price research; 61% for recommendations |
| Potential use cases | Retail purchases, recurring household staples, agent-assisted checkout |
Why this matters now
AI agents are quickly moving from answering questions to making decisions with financial consequences. The moment they can initiate or complete purchases, the industry needs a way to tell the difference between helpful automation and a transaction that never should have happened.
American Express is betting that the answer is to make intent a first-class part of the payment process. If the approach catches on, the next generation of online shopping may be defined not just by how quickly AI can buy something, but by how precisely it can prove that the buyer really meant to.
Frequently asked questions
What is American Express’s Intent Intelligence Framework?
It is a payment framework designed to capture and verify a customer’s purchase intent before an AI agent completes a transaction. American Express says the goal is to make agent-driven purchases safer by ensuring the final charge matches what the consumer authorized.
How does the framework stop unauthorized AI purchases?
It does this by collecting the user’s buying instructions, confirming them with the customer, and then issuing a token tied to those constraints. If the agent tries to buy something outside the approved intent, the transaction can be flagged or blocked before checkout.
Why is American Express building this now?
It is building the system now because AI shopping agents are becoming more common and could scale quickly. American Express says it wants safeguards in place before disputes or fraud become widespread, and consumer surveys suggest shoppers are already open to using AI for shopping tasks.
Will merchants need to change their checkout systems?
Yes, many merchants may need to adapt if they want to support agentic commerce smoothly. The framework is meant to work across the payments ecosystem, but adoption will likely depend on integration by processors, merchants, and other platform partners.
