Overview

Amazon released its Digital Services Act (DSA) transparency report for the first half of 2025, detailing enforcement activity across its European Union marketplaces. The report records 133 million proactive actions taken between January and June, underscoring the scale of automated moderation that sellers now face. Understanding the mechanisms behind these actions and the available appeal pathways is essential for any seller operating on Amazon EU sites.

Key Points

• 133 million proactive actions — The total includes roughly 24 million content removals, 75 million listing disables, and 33 million partial account suspensions across all EU stores.
• 37 000 account‑level actions — These range from temporary holds to permanent bans for sellers whose listings violate EU regulations.
• 172 000 illegal‑content notices — Intellectual‑property claims dominate with more than 127 000 notices, while unsafe or prohibited products generate close to 35 000 alerts.
• Two‑thirds of seller complaints reversed — Out of about 294 000 appeals, Amazon overturned more than 200 000 decisions, with a median resolution time of one day.
• 90 % automated decision accuracy — Only one in ten of the 24 million fully automated actions required manual review or correction.

How Amazon’s EU Enforcement System Works

1. Seller Identity Verification — New EU sellers upload a government‑issued photo ID and business documents. Automated forgery detection and behavioral analytics scan the files, while a cross‑check against a database of flagged actors flags high‑risk applicants. For example, a seller from Spain submitting a passport with mismatched metadata triggers an immediate request for additional proof.
2. Continuous Account Monitoring — After onboarding, Amazon’s system watches for changes in account details, unusual sales patterns, and other risk signals. If a German seller suddenly adds a high‑volume of low‑priced electronics, the platform may pause the new listings and request updated compliance certificates.
3. Algorithmic Catalog Scanning — Thousands of keyword‑based rules and machine‑learning models evaluate billions of product‑detail edits each day. The models detect signs of counterfeiting, IP infringement, or safety hazards. When confidence exceeds a preset threshold, the content is removed automatically; otherwise, it is queued for human review. For instance, a listing that mentions “replica” in the title for a branded watch is instantly disabled.