Policy update · in effect now

Amazon's BSA Agent Policy: What It Means for Sellers and Your AI Tools

On March 4, 2026, Amazon's Business Solutions Agreement gained a new Agent Policy governing AI agents and automated tools. Most coverage stopped at "audit your tools." The more important question is strategic: the rules make scraping-based tools structurally riskier — which is exactly why first-party, listing-level optimization is becoming the durable, compliant play.

By David Daddi · Updated June 2026 · ~9 min read
The short answer

Effective March 4, 2026, Amazon's BSA includes a new Agent Policy: automated tools and AI agents must identify themselves as automated, follow the policy, and stop accessing Amazon on request — plus new restrictions on using Amazon data for AI development. There's no opt-out (continued selling = acceptance), and enforcement is at the seller level, not just the tool. The strategic takeaway: tools built on scraping Amazon data are now structurally exposed, while optimizing your own first-party listing assets — copy, structured attributes, Q&A, images — is the approach least affected by any of this.

What changed

The Agent Policy, in plain English

Amazon announced the update on Seller Central on February 17, 2026; it took effect March 4. Three core requirements apply to any AI agent accessing Amazon's services:

01

Identify as automated

AI agents must clearly declare that they are automated systems, not humans, when accessing Amazon.

02

Comply with the policy

Agents must operate within the requirements of the new Agent Policy attached to the BSA.

03

Stop on request

Agents must be able to immediately cease accessing Amazon's services if Amazon asks them to.

Alongside these, Amazon added AI/ML restrictions on using its materials or services for AI development, with enhanced protection against reverse engineering. And continued use of your seller account after the effective date counts as acceptance — there is no opt-out.

How we got here

This didn't come out of nowhere

The Agent Policy is the contractual endpoint of a posture Amazon has been building for months.

Late 2025
Amazon formalizes its agentic posture

Autonomous tools for advertisers, agentic Seller Assistant capabilities, and a lawsuit against Perplexity over covert AI agent access to the marketplace.

Nov 2025 – Feb 2026
Solution Provider billing tightens

Developers had to submit valid payment info; applications without it lost SP-API access by February 16, 2026 — narrowing who can legitimately access the API.

Feb 17, 2026
BSA update announced

Amazon posts the Agent Policy and AI restrictions to Seller Central, giving sellers roughly two weeks' notice.

Feb 26, 2026
The "data pipeline" reframe

Vanessa Hung (CEO, Online Seller Solutions) argues on LinkedIn that the Agent Policy and Amazon's review-visibility cap are one coordinated move: closing the external data pipeline third-party AI tools relied on.

Mar 4, 2026 →
Agent Policy in effect

The rules are live. Enforcement is at the seller level — Amazon can restrict or suspend the account, not merely block the tool.

The part nobody connected

Why this is a strategy story, not just a compliance chore

Most coverage correctly told sellers to audit their tools. Useful, but it stops one step short. The deeper signal is about which kind of optimization survives.

Vanessa Hung's reframe is the key. She connected two moves that looked separate — Amazon capping visible reviews and the new Agent Policy — and read them as a single intent: closing the data pipeline that flowed out of Amazon for years, feeding third-party AI tools Amazon never authorized or got paid for. If that read is right, the tools most exposed are precisely the ones built on scraping and harvesting Amazon data.

Now contrast that with first-party work. Improving your own listing — the copy you write, the structured attributes you fill, the Q&A you author, the images you upload — isn't automated access to Amazon's services, and it doesn't depend on a data pipeline Amazon is closing. It's the inputs you legitimately own and control. That's not a clever loophole; it's just the part of the game that was never borrowed in the first place.

⚠︎ Honest framing

The "coordinated pipeline closure" is Vanessa Hung's analysis, not an official Amazon statement of intent — treat it as a well-argued read, not a confirmed plan. This page is general information, not legal advice; for your specific tool stack, talk to a qualified advisor. What's solidly true: the Agent Policy is real, in effect, enforced at the seller level, and first-party optimization sits outside its risk surface.

Where the risk actually sits

Exposed vs durable, side by side

More exposed

Tools that scrape Amazon data

Harvesting reviews, BSR, competitor data, or category data — especially to feed models — is squarely in the Agent Policy's sights and the pipeline Amazon appears to be closing.

More exposed

Unidentified automated access

Bots or agents that don't identify themselves, or tools using shared seller credentials, are classic Agent Policy violations waiting to happen.

Durable

First-party listing optimization

Improving your own listing copy, attributes, and Q&A. You own these inputs; optimizing them isn't automated access to Amazon's services.

Durable

Legitimate, identified SP-API use

Authorized, compliant API access for your own account — declared and within policy — remains the sanctioned path for systematic work.

Do this now

Your Agent Policy compliance checklist

Audit before Amazon does it for you

  • List every tool that touches your account — repricers, PPC, listing tools, scripts.
  • For each, answer in one sentence: why does it need access, and what does it do with it? If you can't, cut it.
  • Eliminate shared credentials — give contractors their own limited-permission users.
  • Shut down any internal project scraping product pages, reviews, BSR, or category data to train models — the prohibition is explicit.
  • Keep an internal log of every tool, its permissions, and its compliance status (your evidence if a dispute reaches arbitration).
  • Shift effort toward first-party assets you own and control — the optimization that isn't exposed to any of this.
⚡ 30-second check

How exposed is your current setup?

Four quick questions. No email required.

FAQ

Amazon BSA Agent Policy — straight answers

A policy added to Amazon's Business Solutions Agreement, effective March 4, 2026, governing automated software and AI agents that access Amazon's services. Agents must identify as automated, comply with the policy, and stop accessing Amazon on request. It also restricts using Amazon materials for AI development.

Potentially yes. It covers any automated system touching your account — repricers, PPC software, listing tools, and especially scraping-based tools. Enforcement is at the seller level: Amazon can restrict or suspend the account, not just the tool. Audit every tool that accesses your account or Amazon's APIs.

There's no opt-out. Continued use of your seller account after March 4, 2026 constitutes acceptance of the updated agreement, including the Agent Policy.

The policy formalizes restrictions on unauthorized automated access and on using Amazon data for AI development. Analysts including Vanessa Hung argue it, alongside the review-visibility cap, is a coordinated move to close the external data pipeline third-party AI tools relied on — making scraping-based tools structurally exposed in a way first-party work isn't.

Work on your first-party assets: listing copy, structured backend attributes, Q&A, and images. Improving the inputs you own and control isn't automated access to Amazon's services and doesn't depend on a data pipeline Amazon is closing.

Who wrote this

Why trust Agentic FBA's read?

DD

David Daddi

Founder, Agentic FBA · AI Operator for Amazon · Miami, US

Two areas of expertise that rarely sit in the same person. 25+ years in IT & enterprise architecture since 1999 — including the governance and compliance side of how automated systems access platforms. And a decade operating and teaching Amazon FBA: selling since 2013, a 14,500-subscriber channel, 2,500+ sellers coached, and an FBA incubator that supported 289 startups. Now focused 100% on US brands.

My take: I built Agentic FBA around first-party optimization before this policy landed — not because I predicted it, but because borrowing Amazon's data was always the fragile foundation. The Agent Policy just made the fragility contractual. If your growth depends on tools scraping Amazon, you're renting an edge Amazon can revoke. If it depends on listings you actually own, you're building on ground no one can pull out from under you. That's the whole thesis, and the policy just validated it.

Build on what you own, not what you borrow.

First-party listing optimization is the AI-search play the Agent Policy doesn't touch. See where your own assets stand with the free Scorecard — about 10 minutes, no email required.

Get Your Free Scorecard