Disrupting human trafficking networks using AI on AWS with The Network

The Network estimates that approximately 16,000 illicit massage businesses operate in the United States, which is more than many major fast-food chains have locations nationwide. These businesses generate an estimated 5 billion dollars in annual revenue, and the victims are predominantly trafficked from China and other parts of Asia.

To combat this, The Network maintains a comprehensive database that tracks these businesses and shares its findings with authorities who can take appropriate action. But building a complete picture of each business requires painstaking research: querying multiple data sources; cross-referencing business registrations and property records; and synthesizing information to identify who owns the business, who owns the building, and whether trafficking activity is occurring.

With a team of approximately a dozen people, The Network’s researchers could fully investigate only about a quarter of the businesses in the database each year. The organization needed a way to expand its investigative reach without adding headcount.

The Network’s research requires reviewing disturbing content, including reviews written by individuals who exploit trafficking victims. The organization had previously turned to AWS for help with this challenge. “Reading explicit customer reviews had always had a huge emotional cost for our human researchers,” says Louis Wilbrink, chief technology officer at The Network. “We used AWS to develop an algorithm to read and classify these reviews. After seeing the results, we felt very confident that AWS had the expertise, means, and wherewithal to deliver on a very ambitious project.”

The Network was selected to receive funding and support from the AWS Generative AI Innovation Center—a program that connects organizations with AWS science and strategy experts—to develop an agentic AI research agent. A combined team of six engineers focused on a critical first use case: determining whether businesses in the database are still operating. The agent autonomously performs the same research protocols that human investigators follow, querying multiple sources to verify each business’s status. In less than 4 days, the team created a working proof of concept.

The solution is powered by Amazon Bedrock—used for building generative AI applications and agents at production scale—with Anthropic’s Claude Sonnet 4.5. This foundation provides the reasoning capabilities the agent needs to query multiple data sources, synthesize information, and generate classifications with explanations. AWS offers comprehensive generative AI capabilities with enterprise-grade security, empowering The Network to build sophisticated AI applications rapidly.

The Network adopted AWS Lambda, a service used to run code without thinking about servers or clusters, to provide the serverless compute that powers the research workflow. Amazon API Gateway, a fully managed service that helps create, publish, maintain, monitor, and secure APIs at any scale, is used to manage requests between The Network’s database and the AI system. The agent appears as a widget within the organization’s database, which researchers can use to activate an automated investigation on any business record and receive near real-time updates as the system works through each step.

The Network has taken a measured approach to implementing AI, focusing first on processing publicly available data sources. “We’re very intentional about our AI development road map,” says Wilbrink. “Our current system only analyzes information that’s already public—business registrations, mapping data, social media—while maintaining strict controls around sensitive data handling.”

This careful approach extends to accuracy verification. The Network developed a “golden set” of pre-researched businesses to validate AI outputs against established human research standards. “We’ve built comprehensive validation protocols to verify that our AI agent produces reliable results,” says Wilbrink. “Every automated investigation goes through multiple accuracy checks before the findings are logged in our database.”

The agentic AI solution is currently in the testing phase, with full rollout planned for later in 2026. When the AI agent is incorporated with other tools that are under development, The Network estimates the agent could save thousands of researcher hours annually on routine status verification alone, with AI handling a large percentage of undifferentiated research tasks. This will free The Network’s staff to focus on activities requiring judgment and relationship building, such as identifying business owners and engaging with law enforcement partners.

Beyond increasing efficiency, having AI handle the majority of research tasks reduces the psychological toll on the people working to end human trafficking. “We can save our team from an emotional standpoint—by not having researchers constantly sift through really hard content,” says Meredith Cavan, chief communications officer of The Network. “Having AI triage that content makes a big difference.”

The organization plans to expand its AI capabilities gradually, implementing additional privacy safeguards that align with frameworks like the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act of 1996 (HIPAA) before processing more sensitive data. By maintaining human oversight and limiting AI to clearly defined tasks, The Network has created a framework that both protects privacy and scales its anti-trafficking mission effectively.

“Responsible AI requires collaboration between the builders of AI tooling and models, like AWS, and the implementers of AI—that’s our team and everyone in between,” says Wilbrink. “By delivering gains in cost, effort, and time, AI is helping us advance our mission to fight human trafficking and to eradicate the illicit massage industry.”