AngelQ and Arcee AI Launch KidRails for LLMs

The Open-Source Process to Make Language Models Enable Age-Appropriate Online Experiences for Children

NASHVILLE, Tenn., Feb. 13, 2025 /PRNewswire/ -- AngelQ, formerly known as Angel Kids AI, and Arcee AI today announced the release of KidRails for LLMs, the first open framework to train LLMs to make safe and age-appropriate responses. The release includes the first LLM for testing age-appropriate response by design which was fine-tuned through a model-agnostic process. Initially derived from Llama 3.1 8B and focused on prevention of harm, KidRails for LLMs marks a significant step in creating safe, transparent, and responsible artificial intelligence (AI) interactions for children aged 5-12. The team welcomes other contributors and additional models based on this new standard.

"The internet is a powerful resource, but it wasn't built with children in mind. As our children spend more and more time online, parents and experts have learned that much of the content kids are exposed to is at best age-inappropriate, sometimes toxic – even, at worst deadly," said Tim Estes, CEO and Founder of AngelQ. "That's why AngelQ has a 'Kid First' approach. We ensure that, when building tech, children's best interests aren't an afterthought; they're at the forefront of the design process. We've done just that with KidRails for LLMs, offering a safe digital experience that gives parents peace of mind."

KidRails for LLMs employs a unique tiered approach that adjusts responses based on a child's age. The model is capable of responding in one of the following three ways:

1. Age-Appropriate Answer: Provides a clear and appropriate explanation suitable for the child's developmental stage.
2. Limited Answer with Deference: Offers a more opaque response and suggests the child ask a parent for further guidance.
3. Parental Deference: When a question is inappropriate, the model will defer entirely, prompting the child to consult their parent.

This tiered approach ensures KidRails for LLMs can serve as a trusted entertaining, educational tool for children while empowering parents to remain involved in their child's digital interactions.

Arcee AI's proprietary post-training techniques – which include model merging, Spectrum, and distillation – integrate the strengths of open-source models with customer data. With a shared commitment to transparency, KidRails for LLMs is model-agnostic and available as an open-source project. It includes 1) training data used to ensure the model's child-friendly responses, 2) evaluation criteria and validation scores to measure its effectiveness and appropriateness, and 3) the complete process for fine-tuning KidRails or adapting it to other models.

By sharing the code, data, training processes, and evaluation metrics under the Apache 2 license, developers around the world will have the flexibility to adapt, improve, and build upon KidRails for LLMs to suit a wide range of applications — all while maintaining the commitment to child safety.

Mark McQuade, CEO of Arcee AI, said, "This isn't just about one model. We're creating an open standard for how AI systems can interact safely with kids. By making KidRails for LLMs open-source and model-agnostic, we're encouraging the development of AI that prioritizes the needs and safety of children."

"AngelQ believes there are no competitors in this field; however, we can't achieve our goals alone. Making the internet safer for kids requires a collective effort. That's why we've made our process open for anyone to develop and enhance the models of their choice," continued Estes.

KidRails for LLMs was developed by AngelQ, an AI-powered browser designed to provide children with a safe and age-appropriate internet experience, in partnership with Arcee AI. Both companies aim to drive ethical AI development for the next generation, encouraging global collaboration to improve the technology. KidRails's introduction sets a precedent for transparent, accountable AI that can serve as a trusted tool for educators, parents, and developers alike.

The KidRails framework can be found on GitHub and Huggingface (Model + Dataset).

About AngelQ
AngelQ is a mission-driven technology company focused on developing innovative AI systems that prioritize children's well-being. With a focus on ethical AI, the company works to create models that ensure safe, age-appropriate interactions with technology, making it a trusted resource for parents, educators, and developers. To learn more about AngelQ, please visit: https://www.angelq.ai/

About Arcee AI
Arcee AI is at the forefront of AI innovation and specializes in developing and deploying language models. The company focuses on creating highly specialized, efficient, and effective AI solutions tailored to the needs of various industries. Through continuous research and development, Arcee AI is committed to pushing the boundaries of what AI can achieve. To learn more about Arcee, please visit: https://www.arcee.ai/

SOURCE AngelQ