Security

Is it safe to connect an AI agent to Amazon Seller Central?

Agent Central connects the AI you already use to your Amazon Seller Central account. Yes, connecting it is safe: you approve access on Amazon's own authorization page, so you never share your Amazon password, and your AI client gets a scoped key that you can keep read-only and revoke at any time.

When you do let an agent change your account, change tools return a preview of the current and requested values by default, nothing reaches Amazon without an explicit submit that cannot be applied twice, and each submitted change is recorded in your account history, with the old and new values when it replaces an existing setting.

Sign-in: Amazon's own authorization for Seller Central and Amazon Ads. You never share your Amazon password.
Source: Approved for the Selling Partner API and the Amazon Ads API. No scraping or browser automation.
Keys: Read-only or change-enabled, limited to the areas you choose, and revocable.
Changes: A preview by default, then an explicit submit with an idempotency key.
Audit: Every submitted change is logged, with old and new values for edits to existing settings.

Connection

How does the connection to Amazon work?

You connect Seller Central and Amazon Ads from the Agent Central dashboard. Each one sends you to Amazon's own authorization page, where you sign in and approve the access Amazon lists. Amazon then issues tokens to Agent Central, so you never share your Amazon password and Agent Central never sees or stores it.

Official Amazon APIs: Agent Central is registered with Amazon and approved for the Selling Partner API and the Amazon Ads API. It reads your account through those APIs, never through scraping, screenshots, or browser automation.
Encrypted tokens: Amazon's tokens are encrypted before storage and used only by Agent Central's backend services. They never appear in the dashboard, in tool results, or in your AI client.
Separate accounts: Each account's Amazon data is kept apart from every other account's data.
Revoke at any time: You can revoke Agent Central's access from Amazon whenever you like. Syncs and account changes stop until you reconnect.

Access

What can my AI agent see and do?

Your AI client never receives your Amazon credentials. It connects with an Agent Central key, or with a Connector URL that carries one, and that key decides what the client can reach.

Read-only keys: Pick the read-only preset for an assistant that should only read your data. Any area you add to a key starts read-only, and account-changing tools only work on keys you turn them on for.
Scoped by area: Limit a key to the areas a workflow needs, such as advertising or inventory, or narrow it to individual tools.
OAuth sign-in is read-only: Keys created through the OAuth sign-in flow, which ChatGPT can use, are read-only.
Rotate or revoke: Rotate or revoke any key from the dashboard, and the old key and its Connector URL stop working.

Changes

What guardrails apply before an AI agent changes bids or pauses campaigns?

Every account-changing tool in Agent Central, for advertising, listings, prices, inventory, and shipments, follows the same standard. The guardrails live in the tools themselves, so they apply whichever AI client makes the call. Agent Central returns facts and carries out the changes your agent submits; deciding what to change stays with you and your agent.

Preview by default: Change tools return a preview of the current and requested values unless the call explicitly submits. A preview changes nothing at Amazon.
An explicit submit: Nothing reaches Amazon without an explicit submit, so your AI client can show you the preview and wait for your go-ahead. Account-changing tools are marked as such, so clients like Claude can ask your permission before running one.
Idempotency key: Every submit carries an idempotency key, a one-time ID for that change. If the same change is sent twice, the repeat is caught before it reaches Amazon instead of applying again.
Limits: Bids and budgets stay within bid limits and budget limits for the marketplace currency, large Sponsored Products bid and budget jumps are flagged in the preview, each call has a row cap, and irreversible actions such as archiving need an explicit acknowledgement.
Audit history: Each submitted change is recorded in the Activity tab of your dashboard, with the old and new values when it edits an existing setting such as a bid or budget, so you can check exactly what an agent did.

FAQ

Questions sellers ask about connecting AI.

Is it safe to connect an AI agent to my Amazon Seller Central account?

Yes, when the connection works the way Agent Central's does. You approve access on Amazon's own authorization page, so you never share your Amazon password, and Agent Central reads your account through Amazon's Selling Partner API and Amazon Ads API rather than scraping. Your AI client gets a scoped Agent Central key that can be read-only, and you can revoke the key or Agent Central's Amazon access at any time.

What guardrails should an AI agent have before it changes Amazon bids or pauses campaigns?

It needs a preview of the current and requested values before anything is submitted, your approval before the submit, a duplicate guard such as an idempotency key, bid limits and budget limits, an explicit acknowledgement for irreversible actions like archiving, and an audit history with old and new values. Agent Central's account-changing tools preview by default, require an idempotency key on every submit, enforce the limits and acknowledgements, and log each change, and they are marked as account-changing so your AI client can ask for your approval first.

Is Agent Central safe to use with Amazon Ads and Seller Central?

Yes. Agent Central connects to Amazon Ads and Seller Central through Amazon's own authorization and official APIs, keeps each account's data separate, and gives your AI client scoped keys that can be read-only. When a key is allowed to make changes, each change is previewed by default, submitted only on an explicit call with an idempotency key so it cannot apply twice, and logged in your account history.

Can I give my AI agent read-only access to my Amazon account?

Yes. Create a key with the read-only preset in the Agent Central dashboard, or connect ChatGPT through OAuth sign-in, which creates a read-only key. A read-only key can read reports and live Amazon data but cannot change anything in your account.

Vulnerability reporting

Report a security issue.

Agent Central processes Amazon seller data for authorized customers. If you believe a vulnerability, unauthorized access path, or data exposure affects the service, tell us promptly.

Report

support@agentcentral.to

Use the subject line Security report.

Tracking

We acknowledge the report, open a restricted incident record, assess scope and severity, contain credible risk, and provide material status updates through resolution.

Scope

This process covers the website, dashboard, hosted service, authentication, integrations, and the handling of Amazon seller data.

Include only what we need

A concise description of the issue and its potential impact.
The affected page, endpoint, feature, or integration.
Safe reproduction steps, timestamps, and any relevant request or error identifiers.
Your preferred contact method for follow-up.

Do not send secrets or customer exports

Never send passwords, API keys, OAuth tokens, connector URLs, full customer exports, payment-card data, or other credentials by email. If we need additional evidence, we will arrange a secure, scoped way to collect it.

How we respond

We validate the report and determine whether it affects customer or Amazon information.
We limit exposure by revoking or restricting affected access where needed, then preserve the minimum evidence required for investigation.
We remediate the cause, verify the fix, and record the outcome.
For actual or suspected security incidents involving Amazon information, we follow our Amazon Ads incident-response process and notify Amazon through its required security channel.

Responsible reporting

Do not access, modify, or exfiltrate customer data; disrupt the service; or test beyond what is necessary to demonstrate the issue. We welcome good-faith reports and will work with you to understand and resolve them.

Last updated October 5, 2026