Amazon's Agent Policy: how to keep seller automation and AI tools compliant

Since 4 March 2026, every tool that acts on an Amazon seller account has had a new rule to follow. The Amazon Agent Policy arrived with that month's update to the Business Solutions Agreement. It says automated software and AI agents must identify themselves and must stop when Amazon tells them to. Repricers, scripts, browser extensions and AI assistants all appear to be in scope.

This post is for sellers who run automation on their account and want to know whether it is still allowed, and for anyone planning a new tool. Most coverage so far comes from news and agency sites. I am a developer who builds Seller Central automation, so I will focus on the part they skip: what a tool has to do, technically, to identify itself and be stoppable.

A note on sources. The full Agent Policy sits in Seller Central's help pages, behind a seller login. Where I quote Amazon, I quote its public announcement and its published agreements. Where a detail comes from trade press, I say so. Read the policy in your own Seller Central before relying on anything here.

What the Amazon Agent Policy says

Amazon announced the change on the Seller Forums in February 2026. Updates to the Business Solutions Agreement took effect on 4 March 2026, adding a new Agent Policy with requirements for AI usage and automated systems. According to the announcement, all AI agents must:

1. clearly identify themselves as automated systems;
2. comply with the Agent Policy at all times;
3. cease access if Amazon requests it.

The same update restricts using Amazon materials or services for AI development and strengthens protection against reverse engineering. Carrying on selling after 4 March counts as accepting the new terms.

Amazon's Solution Provider Portal Agreement, which public app developers sign, has a section on agents with matching wording. It defines an agent broadly, as any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, a person or business. It says agents are prohibited from accessing Amazon services once Amazon has explicitly asked them to stop. It adds that Amazon may limit or restrict agent access "through technical or other measures", and it points to the Agent Policy for the technical standards.

What "identify itself" looks like in practice

The seller version of the technical standard sits behind the login. Amazon has, however, published what its agent rules look like in another programme. Its Associates Program Policies, updated 14 April 2026, require an agent to put Agent/[agent name] in the user-agent string of every HTTP request. The agent must also not hide that it is automated, for example by mimicking the speed or pattern of human keystrokes and navigation, or by completing or getting around CAPTCHAs. That programme is not Seller Central, so treat it as a strong indication of the approach rather than the exact seller rule.

For tools built on the Selling Partner API (SP-API), naming the tool is already part of the plumbing. Amazon requires a user-agent header in every SP-API request giving the application's name and version; its example is MySellingTool/2.0. Every call is also made with credentials tied to a registered app. That is my reading as a developer, not a statement from Amazon that SP-API tools are automatically compliant. Still, it is a very different position from a script that drives a browser and passes itself off as a person.

Which tools are affected: repricers, scripts, extensions and assistants

The policy talks about "agents", and the definition is wide. EcomCrew's analysis lists repricing tools, browser extensions used for Seller Central tasks, virtual assistant services that manage seller accounts and "virtually any automation software". Amazon's public announcement does not say where the line sits for simple, routine tools, so the safe assumption is that anything acting on your account without a person clicking is in scope.

Two consequences follow:

• It is your account. The agreement is between Amazon and you, not your tool vendor. EcomCrew advises asking each vendor for written confirmation of compliance and keeping the replies.
• "Kill switch" is the shorthand. Trade press uses it for the stop-on-request rule. In practice it means you, or the tool, must be able to halt all automated access quickly and completely.

SP-API vs scraping and browser bots

Amazon reserves the right to block agents by technical means, as the developer agreement above shows. The tools most exposed to that are the ones that work by pretending to be a person: scripts that log in to Seller Central, browser extensions that click and submit forms, and bots that read pages meant for humans. They also break whenever Amazon changes a page.

For seller automation, the durable route is the official APIs: the SP-API for listings, inventory, orders, reports and pricing, and the Amazon Ads API for advertising. Anything you can only do by scraping Seller Central pages or automating clicks deserves a second look. Either there is an API for it, which is often the case, or it is something Amazon has chosen not to open to software.

If you are running a browser script today, rebuilding it as an SP-API tool is usually the cleanest fix. I explain what the SP-API covers in Amazon Seller Central automation with the SP-API.

How to build automation that identifies itself and can be stopped

Here is a checklist for any tool that acts on a seller account, whether you build it or buy it.

1. List every tool with access

Write down every app, extension, script, virtual assistant and agency login that touches Seller Central or Amazon Ads. Note what each one reads and what it changes. Remove anything nobody can explain.

2. Name the tool in every request

Use the official APIs with a clear user-agent naming the tool and its version. Never disguise automated traffic as a browser, rotate identities or solve CAPTCHAs.

3. Build a kill switch that stops writes

Keep one flag, in a database or configuration store, that every job checks before it changes anything on Amazon. When it is off, queued jobs pause rather than fail. Make sure a non-technical person can flip it. Revoking the app's access or deleting its credentials is the last resort.

4. Respect rate limits

SP-API usage plans are set per operation. When Amazon returns HTTP 429, back off and retry later. A tool that hammers the API is precisely the kind Amazon will restrict.

5. Keep an audit log

Record every change: what changed, the old and new value, which rule triggered it and when. If Amazon ever asks what your automation did, you can answer in minutes.

6. Put a person in front of risky actions

Price drops beyond a set step, listing deletions, refunds and buyer messages should wait for a person to approve them.

7. Keep Amazon data out of model training

If your tool uses an AI model, send it only what each task needs and use an API setting that does not train on your inputs. The new terms restrict using Amazon materials for AI development.

8. Test the stop

Once a quarter, switch the tool off and check that nothing writes to Amazon while it is off.

What Amazon's own AI tools show about expectations

In September 2026 Amazon launched a Selling Partner plugin that connects seller data to Amazon Quick and, in a US beta, Claude. Amazon says it has clear boundaries on what it can access, human approval on actions and complete audit trails. Seller Assistant's always-on workflows log every action too. It is reasonable to read that as the standard Amazon expects from its own agents, and to build yours to match. I compare Amazon's tools with custom ones in Amazon Seller Assistant vs custom tools.

Auditing the tools you already run

Tool type	Ask	Red flag
Repricer	Does it use an official API, and can it pause instantly?	Needs your Seller Central password
Browser extension	Does it click or submit forms for you?	Automates pages, gets around CAPTCHAs
Virtual assistant scripts	Who runs them, and from where?	Shared logins, macros, no log
AI assistant	Which connection does it use?	A browser agent instead of an official plugin

Repricers deserve a closer look, because they change prices without a person in the loop. My post on Amazon repricing automation covers safe guardrails.

What compliance work costs

Most compliance work is small: adding a central off switch, an audit log and a proper user-agent to a tool you already have. Rebuilding a browser script as an SP-API tool is a bigger job, and it fits my marketplace automation MVP, from $2,500: one marketplace connected through its official API and one automation. That is a starting price. Every project gets a fixed quote after a short brief, and the app is registered in your account, so you own it. See the wider e-commerce automation work I do for Amazon, Shopify, Etsy and TikTok Shop sellers.

FAQ

Is the Amazon Agent Policy a ban on automation?

No. Amazon's announcement sets conditions rather than banning tools: automated systems must identify themselves, follow the Agent Policy at all times and stop if Amazon asks. Repricers, inventory tools and AI assistants can keep running if they meet those conditions. What becomes risky is automation that hides what it is, such as scripts that imitate a person in a browser, or tools that cannot be paused quickly when Amazon requests it.

Does a tool built on the SP-API comply automatically?

Not automatically, but it starts in a strong position. Every SP-API request must carry a user-agent header naming the app, and every call uses credentials tied to a registered application, so the traffic is identifiable. You still need to read the Agent Policy in Seller Central, respect rate limits, build a way to stop all writes quickly and keep a log of what the tool changed.

What happens if my automation breaks the Agent Policy?

Amazon's agreements let it limit or restrict an agent's access through technical or other measures, and the announcement says agents must stop when Amazon requests it. Amazon has not published a scale of penalties. Because the agreement is with the seller, the account holder carries the risk, so audit every tool with access and ask vendors to confirm compliance in writing.

If you want your Amazon tools checked or rebuilt on the official API, send me a short brief: which tools touch your account today, what each one does, and which ones you could not live without. Read more about my Amazon automation service, or start with e-commerce automation: what to automate first.