Amazon updated its Business Solutions Agreement (BSA) effective March 4, 2026 to include a new Agent Policy. The policy defines what counts as an "Agent" acting on a seller's account, imposes three baseline obligations on those Agents, and adds restrictions on using Amazon materials for AI training. If you sell on Amazon and use third-party tools, this affects you. Continued selling after March 4 counts as automatic acceptance, with no opt-out.
Amazon announced the change February 17, 2026. It took effect March 4, 2026. Any "Agent" acting on your account must (1) identify itself as automated, (2) comply with the new policy continuously, and (3) cease access if Amazon requests. There is no opt-out, no separate acknowledgment, and continued use of Amazon services after March 4 = automatic acceptance.
The Agent Policy didn't appear in a vacuum. On November 4, 2025, Amazon filed a federal lawsuit against Perplexity in the Northern District of California over Comet, Perplexity's AI browser. The original complaint alleged that Comet disguised its automated sessions as Google Chrome browser traffic and accessed Amazon customer accounts without authorization. Amazon's cause of action was under the federal Computer Fraud and Abuse Act and a California computer fraud statute.
Amazon won a preliminary injunction on March 9, 2026 from U.S. District Judge Maxine Chesney, ordering Perplexity to stop accessing Amazon systems with AI agents and destroy any Amazon data already obtained. Coverage from ppc.land and Retail Dive walks through the timeline.
The signal in the Perplexity case is the spirit of the new Agent Policy. Amazon is willing to pursue legal action against AI agents that act on its platform without proper authorization. The Agent Policy moves that posture from "case-by-case enforcement" to "explicit baseline rules every Agent must follow."
Amazon's definition is broad. An "Agent" includes any automated software or AI system that takes actions on a seller's account or accesses Amazon services on the seller's behalf. Coverage from DAM Law Firm and My Amazon Guy reads the definition as including:
The bar is "does software take actions on your account using your authorization?" If yes, it's an Agent. Read your tools through that lens.
Every Agent must do three things. These are non-negotiable.
The Agent must clearly disclose at all times that it is automated software, not a human. In practice this means proper User-Agent headers in API calls, transparent disclosure when the Agent communicates with buyers, and identifiable signatures in account activity. Agents that try to look like human users (the Perplexity Comet pattern) are explicitly forbidden.
Compliance isn't a one-time check. Agents must continue to operate within the policy throughout their lifecycle. If Amazon updates the policy, Agents must update with it. If an Agent's behavior drifts out of compliance over time, the vendor is responsible for catching and fixing it.
Amazon retains the right to revoke Agent access at any time, for any reason. Agents must have a mechanism to stop operating on a seller's account when told to. This is the most operationally significant requirement: agents that can't be stopped on command are non-compliant.
The Agent Policy isn't the only BSA change effective March 4, 2026. Coverage from EcommerceBytes and EcomClips details several other meaningful changes:
| Obligation area | Before March 4, 2026 | After March 4, 2026 |
|---|---|---|
| Definition of an "Agent" | No explicit Amazon-level definition. Tools were governed by general BSA terms. | Defined: any automated software that takes actions on a seller's account, including third-party AI tools. |
| Self-identification | Not required. Many scrapers used generic User-Agent strings. | Required. The Agent must identify itself as automated on every request. |
| Continuous compliance | Compliance audited mostly when a complaint arose. | The Agent must comply with the Agent Policy at all times, not just at onboarding. |
| Revocation | Process informal. Some tools were hard to remove. | The Agent must cease access immediately when Amazon requests it. |
| Training on Amazon data | Gray area. Several lawsuits open (notably Amazon vs Perplexity, filed November 4, 2025). | Explicit restrictions on using Amazon materials or services to train AI models, with enhanced anti-reverse-engineering language. |
| Mexico BSA | Part of a shared regional document. | Standalone BSA. Sellers operating there must track two agreements. |
| Continued use as acceptance | Not specifically tied to AI tooling. | Continuing to use Amazon services after March 4, 2026 counts as automatic acceptance of the new BSA Agent terms. |
If you haven't done a compliance audit since the policy took effect, here's the practical checklist.
I won't name specific vendors because the compliance landscape is changing fast. What I can say is the categories that got hit hardest:
If you're using something from any of these categories, it's worth a vendor email today.
Quick founder note since I run SellerShorts. The marketplace's architecture was designed before the BSA Agent Policy took effect, but the design choices we made map cleanly to the new requirements.
This isn't a sales pitch. It's the answer to "what does BSA Agent Policy compliance look like in practice for a marketplace platform." If your other tool vendors can't describe their own answer in similar terms, that's a flag.
Amazon's enforcement options range from soft to hard:
Most sellers won't hit hard enforcement quickly. Amazon tends to start soft. But the policy is real, and the trajectory is enforcement that becomes more automated and more strict over time.
Amazon updated its Business Solutions Agreement on March 4, 2026 to include an Agent Policy that defines what counts as an 'Agent' acting on a seller's account and imposes three baseline obligations: identify as automated, comply continuously with the policy, and cease access on Amazon's request. Continued selling after March 4 counts as automatic acceptance with no opt-out.
Any automated software or AI system that takes actions on a seller's account or accesses Amazon services on the seller's behalf. That includes repricers, PPC managers, listing tools, inventory systems, customer-service bots, AI agents from marketplaces like SellerShorts, and even simple scheduled SP-API scripts. The bar is whether software acts on your account using your authorization.
No explicit acceptance was required. Continued selling on Amazon after March 4, 2026 counts as automatic acceptance. There was no opt-in form or separate acknowledgment.
On November 4, 2025 Amazon filed suit against Perplexity in the Northern District of California over Comet, Perplexity's AI browser. Amazon alleged Comet disguised automated sessions as Chrome browser traffic and accessed Amazon accounts without authorization. Amazon won a preliminary injunction on March 9, 2026. The case set the stage for the broader BSA Agent Policy by moving Amazon from case-by-case enforcement to explicit baseline rules every Agent must follow.
No. The Amazon Ads MCP Server launched February 2, 2026, before the BSA Agent Policy took effect. Any agent using MCP to act on your account is still an Agent under the BSA definition. The MCP layer does not exempt the agent from the three baseline obligations.
SellerShorts is built on capability-scoped OAuth, identifies as automated to Amazon, and supports revocation at any time. AI Tool builders publish their tools on top of this BSA-aware foundation, so any tool published inherits the compliance posture. If you need a compliant tool for a use case that isn't published yet, post a request and AI Tool builders see the demand.