Since March 4, every automated tool that touches your Amazon account must identify itself, follow the Agent Policy and stop when Amazon says stop. This guide covers which tools are in scope, how enforcement works and the audit DTC brands should run now.
Since March 4, the repricer that runs overnight on your account counts as an agent, and Amazon holds you responsible for what it does. The same applies to your bid optimizer, your review-request tool and the forecasting app that pulls your FBA reports. Amazon's AI agent policy now sits inside the Business Solutions Agreement, so any automation that breaks it puts your seller account at risk. Most brands accepted the update without noticing, because continuing to sell counted as acceptance. This guide explains how to find out whether your tools comply.
What the Agent Policy actually says
Amazon posted the BSA update to the Seller Central forums on February 17, 2026, which gave sellers two weeks' notice. The official announcement adds a standalone Agent Policy for "automated software or AI agents" that access Amazon Services. It sets three obligations. Every agent must clearly identify itself as an automated system, comply with the Agent Policy at all times and stop its access if Amazon asks. Amazon also says it may restrict agent access "in certain instances", and it gives no threshold for when that happens.
A second change matters if you build anything in-house. The BSA now restricts using Amazon materials or services for AI development and adds protection against reverse engineering.
The bigger change is who carries the risk. The Agent Policy is part of the Business Solutions Agreement, your own contract with Amazon, so your vendor's terms of service do not shift any of that responsibility onto the vendor. When a vendor's tool misbehaves, the problem lands on your account, not the vendor's.
Be wary of how this policy has been retold. Many blog posts quote specific rules, such as 12-month audit logs or human sign-off on price changes above a set percentage. Those details do not appear in Amazon's announcement. Open the Agent Policy in Seller Central Help and work from Amazon's own wording.
Automation without account risk
We audit and run compliant Amazon automation stacks
Hyperzon checks every repricer, bid tool and messaging app on your account against the Agent Policy.
More than most sellers think. In legal readings of the policy, any software that takes actions or pulls data automatically using your authorization is an agent. A weekly report pull counts the same as a repricer that runs every few minutes. How often a tool runs affects your practical risk, not whether the rules apply.
Pricing
Repricers such as Aura, Seller Snap and BQool write prices to your offers with no human approving each change. They are agents. You also own the outcome when an algorithm drops your price below your margin floor or pushes it into a range Amazon flags as unfair pricing.
Inventory management and forecasting
Forecasting and replenishment tools pull inventory and sales reports, and many of them create inbound shipment plans. Pulling data counts. A forecasting system that only reads your account is still an agent, and how the vendor uses that data matters under the AI development restriction.
Customer messaging and review requests
Review-request software and AI customer service bots act on buyer data, which makes them the most sensitive category. Amazon's announcement does not require you to tell buyers that a message came from AI. The self-identification duty is owed to Amazon. Your messages still have to follow Amazon's communication rules, and the safest set-up for compliant review requests is still the official Request a Review flow.
Listing updates
AI copywriting tools that push titles, bullets or attributes to your listings are agents. Browser extensions that edit listings by clicking through the Seller Central interface carry the most risk. They log in as you and imitate a human user, which conflicts with the self-identification requirement.
Advertising bid management
Tools such as Perpetua, Pacvue and Teikametrics connect through the Amazon Ads API, not the selling account. PPC Land's coverage notes that sponsored ads management tools could fall under the definition depending on how Amazon enforces it. Treat them as in scope. It costs you little, and an unexplained bid spike is exactly the kind of behaviour you will need to account for.
The behaviour Amazon is targeting is easy to spot. Amazon sued Perplexity on November 4, 2025 over a Comet browser agent that allegedly presented itself as Google Chrome. Any tool in your stack that hides the fact that it is automated works the same way.
Registration and disclosure inside Seller Central
Many operators look for an "AI agent registration" form in Seller Central. The announcement does not describe one. Disclosure happens at the authorization layer.
It works like this. The software vendor registers its application with Amazon through the Solution Provider Portal, which the BSA update renamed from "Developer Site". You authorize that application, and it appears in Seller Central under Apps and Services > Manage Your Apps. That page is your official record of every agent with authorized access, and it is also where you revoke access. Ads tools are authorized through your Amazon Ads login instead, so keep a separate list of those connections.
In short, a compliant agent is a registered application that you can see and switch off. Tools that use your password, a shared VA login or a browser session do not meet that standard.
How Amazon enforces the policy
Amazon has two levels of enforcement. At the tool level, it can restrict an agent's access or tell it to stop. If the tool cannot stop on command, you are in breach.
At the account level, the Agent Policy is part of the BSA, so a breach is a contract violation and not a performance metric problem. That puts your selling account at risk, and a deactivation notice may not point to the tool that caused it. Disbursements freeze while you work out the cause.
Appeals are harder than usual. A general statement that you comply will not work. You have to show that the specific tool was fixed or replaced. Read our suspension plan of action guide before you need it, and add tool changes to your account health monitoring routine so you can match any warning to a recent automation update.
The compliance audit checklist for DTC tool stacks
Run this audit once in full, then again every quarter and after every major vendor release. Tools change through updates, so one check at setup is not enough.
Build the full inventory: Export Manage Your Apps and list every Ads API connection. Then add the tools that do not appear there, such as browser extensions on team laptops, scripts run by VAs and anything that uses a shared login. Unknown apps on the list are either forgotten tools or unauthorized access, and both need action today.
Classify by action: Mark each tool as read-only or write. Then record what it writes: prices, listing content, bids, budgets, buyer messages or shipments. Tools that write carry the most risk.
Verify the connection method: Confirm that each tool uses a registered SP-API or Ads API application. Retire any tool that works by logging in as you.
Test the kill switch: Revoke a tool's access in a quiet hour and check that its activity actually stops. If you cannot cut a tool off within minutes, it fails the "cease access" requirement.
Audit repricer guardrails: Check that every SKU has a minimum and maximum price, along with alerts for unusual price changes. An algorithm with no floor is a liability.
Audit review-request logic: Remove any conditional sending, such as only asking buyers who left positive feedback, and send requests through Amazon's official solicitation flow.
Audit bid optimizer limits: Set caps on bids and daily budgets, and make sure the tool keeps a change log. If you outsource PPC management, the same rules apply to your agency's software.
Check how forecasting tools use your data: Ask whether your Amazon data trains a shared model across the vendor's clients. That is the AI development question to settle with each vendor and, where needed, with counsel.
Revoke what you cannot verify: Turn off access for any tool you cannot confirm while you wait for the vendor's answers. Revoking access pauses the tool without cancelling the subscription.
Questions to put to every vendor
A vendor's statement that it is "compliant" does not move responsibility back to the vendor. Ask for specific answers you can verify:
Connection: Do you connect only through a registered SP-API or Ads API application? What is your application ID, and does it match the entry in my Manage Your Apps list?
Browser automation: Does any feature, including Chrome extensions, act in Seller Central through a browser session instead of the API?
Stopping: If Amazon tells you to stop, how fast do all actions on my account stop, and how will you let me know?
Logging: Can I export a record of every action the tool took on my account, with timestamps and the rule or model that triggered it?
Data use: Is my Amazon data used to train or improve models that other clients use?
Change notice: Will you tell me before releases that change what the tool does automatically?
Get the answers in writing. A vendor that will not answer the data-use question has already told you what you need to know.
Document it as if you will need it for an appeal
If Amazon ever questions a tool, you will need written records. Keep an Agent Register with one row per tool. Record the vendor, the application ID, the connection method, what the tool reads and writes, who owns it internally, the human controls in place, the date of the last kill-switch test and a link to the vendor's written answers.
Add a monthly export of action logs from every tool that writes to your account, starting with repricers and bid optimizers. If you get a deactivation notice that does not name the cause, these records let you check which tools changed recently and show Amazon exactly what you fixed. The same records also help when you hand over inventory management or onboard a new team member.
This week, export your Manage Your Apps list, compare it with what your team actually uses, and revoke everything nobody can explain.
Under the Agent Policy, every automated tool on your account is your responsibility, and the safest fix is a stack you can audit, log and switch off. Hyperzon runs the audit, replaces tools that fail it and manages pricing, PPC and inventory automation with documentation ready for an appeal.Get a free Amazon audit.
Article was originally published on 06 November, 2026
Frequently Asked Questions
What does Amazon's Agent Policy require from sellers?
Amazon's announcement lists three obligations for every AI agent or automated tool that accesses Amazon Services. It must clearly identify itself as an automated system, comply with the Agent Policy at all times and stop its access if Amazon asks. Amazon also reserves the right to restrict agent access, and the updated BSA limits using Amazon materials or services for AI development.
Is there an AI agent registration form in Seller Central?
Amazon's announcement does not describe a separate seller-facing registration form. In practice, a tool identifies itself through its registered developer application. You authorize that application, and it then appears in Seller Central under Apps and Services, Manage Your Apps, where you can review or revoke it.
Do repricers, PPC tools and inventory software count as agents?
Legal analysis of the policy treats any software that acts or pulls data automatically using your authorization as an agent. That includes repricers, bid optimizers, listing tools, inventory systems, reimbursement services and customer service bots. How often a tool runs makes no difference.
What happens if one of my tools breaks the Agent Policy?
Amazon can restrict or end the tool's access. Because the policy sits inside the Business Solutions Agreement, a serious breach can also put your selling account at risk, and an appeal has to show that the specific tool was fixed or removed.
Do I have to tell buyers that a message was written by AI?
Amazon's announcement frames self-identification as something agents owe Amazon. It says nothing about labelling messages for buyers. Every automated buyer message still has to follow Amazon's communication guidelines, and review requests should go through the official Request a Review flow.
Can I use my Seller Central data to train my own AI model?
Be careful. The updated BSA restricts using Amazon materials or services for AI development and adds protection against reverse engineering. Get legal advice before you export account data into a custom model, and ask vendors whether they train shared models on your data.
Radoslav Mazganov
VP Hyperzon by Moburst
With over 10 years of experience in e-commerce and Amazon advertising, Radoslav helps brands grow through expert PPC strategy and data-driven optimization. His deep platform knowledge drives visibility, performance, and scalable results.