CPSC Should Leverage AI to Modernize Product Safety
The Consumer Product Safety Commission (CPSC) protects the public from hazardous products by setting safety standards, flagging dangerous goods, and issuing recalls. However, its enforcement model today depends largely on consumer complaints—a reactive approach that leaves gaps in protection. To modernize enforcement, CPSC should use artificial intelligence (AI) to analyze real-time and historical data, allowing it to predict and address risks in e-commerce supply chains before harm occurs.
Two challenges make this modernization urgent. First, the shift to e-commerce has enabled vendors to bring foreign goods to the United States at unprecedented speed and scale. Second, popular Chinese e-commerce platforms like Temu, SHEIN, and AliExpress have poor product safety track records. For example, testing has found that shoes sold on these platforms contain unsafe levels of toxic substances. In September 2024, CPSC called for an investigation into Chinese e-commerce platforms’ safety controls, but their growing popularity highlights the need for stronger detection capabilities now.
To maximize AI’s impact, CPSC should focus on three areas: Enhancing current monitoring, enabling real-time detection, and future-proofing enforcement against evolving tactics. This requires technologies ranging from established tools like reverse image search to sophisticated generative AI models that autonomously detect anomalies across vast datasets.
For enhanced monitoring, CPSC should implement sentiment analysis software to track consumer feedback across platforms. This technology already exists in commercial applications like Fakespot and Helium 10’s Amazon Review Checker. If negative sentiment suddenly spikes around a children’s toy with complaints like “caused a rash,” CPSC could receive automated alerts and initiate investigations before formal complaints are filed. Additionally, reverse image search capabilities would help implement recalls across platforms by matching official product photos with resellers’ listings.
For real-time detection, CPSC needs machine learning models for anomaly identification. These algorithms could rank unusual patterns in product sales, returns, and reviews to generate leads for CPSC analysts. By flagging anomalies like sudden surges in suspiciously low-priced items, inconsistent seller metadata, or high return rates with quality complaints, the system could identify dangerous goods before widespread distribution.
Such models require high-quality training data. CPSC would need datasets from commercial brokers, public-private partnerships with e-commerce platforms for anonymized sales data, and integration with government sources like Customs and Border Protection (CBP) shipment records. Training classifiers on historical counterfeit data would improve detection accuracy over time.
Bad actors often distribute counterfeit goods through complex, multi-tiered supply chains involving multiple online storefronts, resellers, and international manufacturers. CPSC should not only flag product safety issues when or before they create consumer harm but also work to proactively identify counterfeiting organizations further upstream. Through interagency collaboration with CBP, CPSC could deploy graph-based network analysis to trace illicit networks, identifying repeat offenders and visualizing connections between seemingly unrelated sellers. This analysis might reveal multiple sellers operating under different brand names but sharing common IP addresses, payment processors, or warehouse locations.
Creating an AI-centric CPSC would not be easy. CPSC would need specialized talent who understand both CPSC’s needs and how to automate software to fit those needs. To obtain the necessary data to train future models, CPSC would need to address legal issues relating to data-sharing frameworks. However, CPSC already has two advantages.
First, it would not start from scratch. Beyond data—which is one of CPSC’s most valuable assets—and algorithms, the success of AI tools depends upon methodological questions of measuring success and defining metrics. While the agency's analytical work largely occurs after hazards are reported rather than in proactive identification, it has a strong track record of creating robust methodologies used by other agencies, such as the National Electronic Injury Surveillance System, a globally used public health research tool. Second, CPSC is already institutionally interested in AI; in its recently revised 2023-2026 Strategic Plan, it pledged to utilize AI to broaden its analysis of hazard data and better identify emerging hazards. CPSC also enjoys bipartisan support for the implementation of AI for its mission.
As e-commerce continues to evolve and risks multiply, CPSC should adopt advanced AI tools to keep pace with emerging threats and ensure consumer safety. These tools are increasingly feasible and accessible for government implementation, offering powerful means of safeguarding the public in real time rather than reacting after harm occurs.
Image credit for social media preview: Generated by DALL-E
