Amazon Uses AI to Seize 15 Million Counterfeits

What happened - Amazon released its first Trustworthy Shopping Experience Report and reported it identified, seized, and disposed of more than 15 million counterfeit products in 2025, blocked hundreds of millions of suspected fake reviews, and has pursued over 32,000 bad actors since 2020. The company says these results come from a blended strategy of AI-driven predictive detection, new physical inspection hardware, seller onboarding controls, and targeted legal enforcement.

Technical details - Amazon credits automated, predictive systems that process large-scale signals to find fraud early. Key capabilities cited include: - Project PI, an imaging tunnel system deployed in fulfillment centers that uses cameras and AI to detect cracks, dents, and expired dates as items move through the line - Predictive listing and review classifiers that flagged coordinated fake-review campaigns and stopped listings before brands reported them - Expanded seller verification and onboarding controls requiring proof before new sellers can list - The Transparency programme, which has verified 2.7 billion product units and covers over 90,000 brands - Automated monitoring of more than 90 million customer interactions per week to surface anomalies and phishing URLs

Amazon pairs these technical controls with enforcement actions: takedowns of more than 100 fake-review websites, legal suits and criminal referrals targeting organized networks across 14 countries, and increased phishing URL takedowns and call-blocking for scam impersonations.

Context and significance - This report is a scale play. Amazon operates a marketplace with billions of listings and millions of transactions, so incremental gains in early detection directly reduce harm at scale. The shift to predictive AI detection matters because it changes the problem from reactive takedown to proactive disruption of counterfeit economics. For practitioners, the notable signals are the marriage of computer vision at fulfillment centers with large-scale fraud telemetry models that ingest reviews, complaints, seller history, and customer-service interactions.

The strategy mirrors trends in other sectors where AI models provide early-warning signals and automation handles triage, while legal and human teams pursue higher-cost enforcement. Amazon's work also underscores the operational complexity of deploying vision systems in fulfillment hardware and integrating those outputs into supply-chain workflows without introducing false positives that disrupt legitimate sellers.

What to watch - Watch for independent evaluation of false positive rates from Project PI and predictive classifiers, how Amazon balances seller friction with fraud prevention, and whether competitors adopt similar fulfillment-center imaging. Expect continued growth in cross-border enforcement actions as Amazon pushes for international cooperation to dismantle organized counterfeit networks.

Why it matters for practitioners - The report shows a concrete example of productionizing multilayered AI for fraud detection: fusion of vision models, long-running telemetry classifiers, and policy workflows. Teams building marketplace security systems should study the integration patterns here: early detection models feeding automated fulfillment gates, paired with legal escalation paths to change attacker economics.

Bottom line - Amazon demonstrates that combining AI with physical inspection and legal pressure can materially disrupt large-scale counterfeit operations. The remaining open questions are model performance in deployment, seller impact, and whether predictive approaches scale across less-resourced marketplaces.