New Laws Tighten AI Guardrails for Minors

- Oregon's SB 1546 specifically forbids AI companions from using reward systems to maximize a minor's engagement time or generating messages of emotional distress if a user wants to end a conversation. The bill also requires hourly reminders for minors to take a break from the chatbot interaction. - The Colorado bill, SB26-051, proposes that a device's operating system would collect a user's age to generate an "age bracket" signal, which would then be made available to app developers through an API. This approach centralizes age verification at the OS level, rather than requiring each app to implement its own mechanism. - In Germany, the Social Democratic Party has proposed a three-tiered system: a complete ban on social media for children under 14, a mandatory "youth version" with no algorithmic feeds for those under 16, and deactivating algorithmic recommendations by default for everyone 16 and older. - For edtech applications, machine learning algorithms can create personalized learning paths by analyzing student interaction data, assessment scores, and feedback to identify what is and isn't working for similar student clusters. - Research into AI reading tutors shows they can provide immediate, individualized feedback on phonics and pronunciation, which is difficult for a single teacher to offer in a large classroom. However, studies have also found that reading with a parent yields better listening comprehension outcomes compared to reading with an AI chatbot. - Federal laws impacting AI in education include COPPA, which has updated rules as of June 2025 for tools used by children under 13, and FERPA, which requires written agreements that prevent vendors from using student data to train their own AI models. - A key challenge in developing adaptive learning systems is the need for large amounts of high-quality, digitized learning data to train the ML models effectively, which can raise data privacy concerns under regulations like COPPA and FERPA. - Studies on AI tutoring effectiveness show that unrestricted access to an AI tutor can be more beneficial than restricted access, leading to higher test performance by fostering a more gradual integration of AI support. These benefits are most pronounced for students with lower baseline knowledge but strong self-regulation skills.