NIST Ranks Top Age Estimation AI
In a new evaluation focused on child online safety, NIST has ranked AI provider ROC as #1 for age estimation accuracy. The company's algorithms performed best in both child online safety and mugshot datasets, setting a new benchmark for biometric precision.
The National Institute of Standards and Technology's (NIST) evaluation is part of the Face Analysis Technology Evaluation (FATE), a rebrand of the former Face Recognition Vendor Test (FRVT). FATE focuses specifically on algorithms that analyze facial images for characteristics like age, while identity verification now falls under the Face Recognition Technology Evaluation (FRTE). Performance is primarily measured by Mean Absolute Error (MAE), the average difference between the AI's age estimate and the person's actual age. In a recent FATE report, ROC, the top US provider, achieved an average MAE of 2.96 years across application, border, and mugshot datasets. For context, other top-performing vendors in various NIST evaluations include Innovatrics, Yoti, and Regula. The push for this technology is driven by legislation aimed at protecting minors online, such as the Children's Online Privacy Protection Act (COPPA). In a significant policy shift, the Federal Trade Commission (FTC) announced in February 2026 that it would not pursue enforcement actions against companies using this tech for the sole purpose of age verification, encouraging its adoption. These evaluations are critical for scenarios like the "Challenge 25" test, designed to prevent minors from accessing age-restricted content by flagging if a user appears under 25, which then triggers further verification. This requires a low False Positive Rate (FPR), meaning the algorithm rarely misidentifies a minor as being over the threshold. To ensure robustness and minimize bias, NIST tests these algorithms against a massive and diverse collection of images. Past evaluations have used datasets containing millions of photos, including mugshots, border-crossing webcam images, and immigration photos of people from over 100 countries.
