Our Story

Online content offers a variety of educational, social, and entertainment benefits to kids and teens. However, a considerable amount of harmful content is also transmitted directly to kids’ phones. This content can negatively impact a child’s behavior and compromise their safety. A recent analysis of over 1 million messages collected from teens and tweens demonstrated that 8% of their online conversations contained potentially harmful content. Among the alerts found, 37% were release of personally identifiable information (name, phone, email, address), 16% contained inappropriate content (sex, drugs, pornography), 15% were mental health related, and 14% were contacts with an adult stranger.

boundrees is a next generation content-monitoring solution that automatically captures text-based communication and alerts parents of harmful conversations. The algorithm that underlies the boundrees app detects eight categories of harmful content: cyberbullying, grooming, sextortion, mental health issues, drug/alcohol use, exposure to inappropriate content, release of personally identifiable information, and, violence, including dating violence. Distinctive to boundrees, the content monitoring and alert system is designed to run entirely on a mobile device such as a cell phone. This approach greatly reduces the risk of data leaks and maintains the child’s privacy in the absence of concerning behavior. The boundrees app will only transmit data to the parent’s device when a harmful event is detected. At that time, the parent will receive a notification with the problematic message and additional context (surrounding messages) in the parent companion app.

Boundrees was developed from over 12 years of NSF-funded research on how teens and tweens communicate online. The heart of boundrees is a light-weight artificial intelligence algorithm that is fast and highly accurate. It uses outbound communication from the child’s phone to detect harmful events in real-time, regardless of the app being used. The boundrees app is fully compliant with Apple and Google Play terms and conditions, thus there is no need to “jail break” or “side load” the app. You can get it directly from The App Store and Play Store.

In recent benchmark testing, the AI algorithm at the heart of boundrees achieved 95% accuracy, detecting 232 harmful events in milliseconds, while ignoring harmless messages. Our algorithm is not based on specific keywords, but instead uses the context of a communication stream, capturing real-time information on what kids are saying, and how they are reacting to incoming messages.

The boundrees AI was trained using over 1 million labeled messages from real kids (ages 8–14) who participated in our data collection studies. These messages were manually labeled by a data science expert, using a structured protocol. The kids who participated in our study collectively used over 900 unique apps over the course of a year (not websites, actual downloaded apps). Importantly, the kids knew that they were being monitored and the data was being collected. Prior research has shown that kids don’t change their behavior even when they are aware that they are being monitored. A few major incidents (and many minor incidents) were detected during the labeling process. The boundrees team worked closely with parents to protect kids who were being groomed by adults they did not know, and, in one case, prevented a significant violent incident.

The boundrees team is committed to ongoing research in online communication and regularly updates the AI engine. We will periodically offer opportunities for families who are interested in participating in these studies. Watch your email for more information!