Shopify opens its catalogue to AI agents, and the *receipt question* follows

The catalogue opens

Shopify announced on 10 October 2026 that it is opening its product catalogue to AI agents ^[1]. Any agent can browse the catalogue, and consumers can let their agents buy things, with approval rules they set themselves. The catalogue holds more than a billion products.

The company says every purchase needs consumer approval. That could mean a confirmation for every item, or a standing instruction like "buy coffee beans when I'm running low, up to 500 kronor a month". Either way, the agent acts, and someone has to be able to show what it did.

The receipt question

When you buy something yourself, you get a receipt. When an agent buys it for you, the receipt question is trickier: can you show what the agent bought, when it bought it and on whose instruction?

Say your agent orders a coffee grinder and you get charged 4,000 kronor instead of 400. You check the order history and see the expensive model. Your agent's provider says the agent followed your standing instructions. You say you never asked for that model. Without a checkable record of what the agent was told and what it did, the argument is your word against the provider's logs.

That is the everyday version of the evidence question.

What the AI Act says

From 2 August 2026, people must as a rule be told when they are interacting with an AI system ^[2]. The requirement is in Article 50 of the AI Act. It applies to AI that communicates with people, unless it is obvious from the circumstances that the system is an AI, or it is authorised by law to detect or prevent criminal offences.

For an agent acting in a shop, the first question is whether the consumer knows an AI is doing the browsing and buying. Shopify's announcement suggests the agents identify themselves, which would meet that rule.

The second question is the one Article 50 does not answer: who keeps the record of what the agent did, and can you check it?

How a witness works here

A digital witness fits this case well. The thing to seal is each step the agent takes: the instruction it received, each tool call it made ("browse catalogue", "add to basket", "confirm purchase") and the result it got back.

Sealing takes a fingerprint of the record, not the record itself, so the agent's actions and your instructions stay with you and the provider. The fingerprint goes to WitniumChain, and you get a witness ID back. When there is a question—say, six months later—you calculate the fingerprint again and check it against the sealed one. If they match, the record has not changed since the agent acted. Change a single character and it will not match.

Anyone you share the record with can do the same check. That means you, the provider, the shop and, if it comes to it, a court can all see the same thing and agree on what happened, without having to trust whoever kept the log.

How it would work

1. The agent receives an instruction, calls a tool and logs the result.
2. Your system sends a fingerprint of that log entry to WitniumChain and gets a witness ID.
3. Later, anyone calculates the fingerprint again and checks it against the sealed one.

A witness does not authorise the purchase or show that the agent made the right choice. It shows that the record of what happened is the one written at the time, unchanged.

You can seal agent actions through the MCP server, the SDK or the REST API. The content stays with you. The first 2,000 witnesses a month are free. Read the docs.

What this means for you

If you are building an AI agent that acts on someone's behalf, here are three questions worth asking now:

• Can you show what your agent did, step by step, in a way the user can check?
• If the user disputes an action, whose record decides what happened—yours, or a screenshot from the user?
• Are you sealing instructions, tool calls and results as the agent acts, so both sides have the same checkable record?

For anyone letting an agent shop for them, the question is simpler: does the agent give you a receipt you can trust?

What to watch

Article 50's transparency requirement applies from 2 August 2026 ^[2]. Shopify's move suggests agent-based shopping will be common before then.

The general-purpose AI code of practice, expected in early 2025, may say more about logging and transparency for agents, though it will not be binding law.

If agents start acting widely without checkable records, expect the first disputes to arrive quickly, and the first regulatory questions to follow.