How Amazon is experimenting with artificial intelligence to advance human rights

AI pilot projects are helping us predict, prevent, and respond to human rights risks around the world.

Amazon is committed to respecting the human rights of all people connected to our business, and we’re always innovating to improve our ability to identify, prevent, and mitigate human rights risks. That includes exploring the use of machine learning and artificial intelligence (AI) to optimize and scale social audits—regular on-site checks of our third-party suppliers’ facilities to assess human rights and environmental issues like worker health and safety—and building AI-driven models to help us better predict risk in our value chain.

"The vast scope of today’s global supply chains requires new tools to effectively target human right risks," said Leigh Anne DeWine, Amazon’s director of human rights and social impact. "With a global network of hundreds of thousands of suppliers, we're developing AI tools to enhance—not replace—human judgment in upholding our standards. We are experimenting with machine learning and AI to help transform massive amounts of data into actionable insights, enabling more effective risk assessment, audit processing, and decision-making across our complex supplier network."
 

Amazon is still evaluating and improving these tools, but early results are promising. 

Pre-audit: smart risk prediction

We developed an AI model that analyzes data from tens of thousands of historical social audits to identify risk patterns and flag whether suppliers are likely to meet Amazon’s Supply Chain Standards, our code of conduct for all third-party suppliers. This allows us to focus our auditing resources on higher-risk suppliers.

The AI system works in three ways: analyzing existing historical data, enhancing that information with computer-generated data, and looking for new emerging risks.

When we have extensive historical data available, our AI system analyzes past audit information to assess human rights risks. Think of it like airport security: the AI system acts like a scanner that flags potentially dangerous items. Humans reviewing the system act like officers who double-check alerts, making final decisions and catching anything the scanner might have missed. We keep track of the system’s recall (how many threats it catches), its precision (how often it flags something), and its overall accuracy. In production testing, the AI model currently achieves 90% recall, 62% precision, and 85% overall accuracy, correctly flagging approximately nine out of every 10 high-risk supplier sites while keeping false alarms low.

When historical data is limited, we use computer-generated simulated examples that mirror real-world situations. This approach helps our system learn to identify potential human rights risks even in cases where we don't have previous audit data.

Finally, the system uses natural language processing and real-time monitoring to scan global media, government reports, and macroeconomic indicators to identify potential emerging risks, such as worker complaints or factory shutdowns.

We’ve worked to ensure the system accurately interprets risk in the context of a supplier’s region and industry sector, and we’ve built tools to help human rights risk managers understand why the system may have flagged a supplier for potential human rights risks.

Post-audit: accelerated intelligence

Human rights managers review thousands of audit reports each year; reading and analyzing each report manually can take up to four hours per audit. We've developed a new generative AI tool to review these reports in minutes, identifying risks, checking them against our Supply Chain Standards, rating how serious they are, and then suggesting next steps. An early version of this tool helped us process audit reports 65% faster, allowing our teams to quickly identify where action is needed.
 

This tool can also quickly identify high-priority issues like child labor risks and safety concerns. Every AI recommendation is reviewed by our experts, which helps us improve both accuracy and model performance over time.

The road ahead
 

Built using Amazon Web Services (AWS) technology, these AI initiatives mark the beginning of our journey to integrate artificial intelligence into our human rights work.

"AI is fundamentally changing how we approach responsible business—giving us the capability to identify risks with new levels of accuracy and speed," said Kommy Weldemariam, chief scientist for sustainability and AI at Amazon.

As we continue developing these AI tools, we face two key challenges: gathering more comprehensive data to train our AI systems and encouraging wider adoption of these technologies across the industry. That's why we're collaborating with audit firms, non-governmental organizations, and research organizations to improve data coverage for the AI modeling and rigorously benchmark the model performance. Together, we can better understand conditions across our global supplier networks and share what we learn to scale responsible practices industry-wide.

Learn more about how Amazon is advancing human rights.

Sign up for our newsletter to get Amazon sustainability updates sent directly to your inbox.