Our Data Governance Principles

Our Data Governance Principles

Learning Commons builds public AI infrastructure for education: structured educational datasets, evaluation tools, and technical infrastructure for the development of AI tools that better support learning experiences.

Building that infrastructure responsibly means being deliberate about the information AI systems actually need. AI products may rely on a variety of context and inputs to do their work, but our approach is to use only what is necessary to support teaching and learning tasks.

That principle shapes how we build Learning Commons tools, especially as the products and AI systems leveraging our tools continue to evolve. We limit the information our services need, protect the information we do receive, and give developers choices over what they share.

For the specifics of how we collect, use, and protect information, you can also review our Privacy Policy and Security Overview.

Built to work without student data

Much of what Learning Commons provides is educational context from third-party educational sources: information such as state academic standards, curricula, learning progressions, and other learning science datasets and rubrics about what students learn and how their learning progresses.

A developer querying Knowledge Graph, for example, might ask about the skills associated with a particular academic standard, the learning components appropriate for a grade, or the concepts that come before and after a particular skill. Learning Commons tools need only enough information to answer the educational questions, meaning the tools are built so they do not utilize a student’s personal information or work.

When schools, districts, and developers use these resources, student records stay inside their systems — such as their learning management systems, gradebooks, or AI products. Learning Commons’ role is limited to supplying high-quality educational context (state academic standards, skills, learning progressions, etc.) to the tool builders. That distinction matters to us because the goal of our infrastructure is to make educational expertise more useful to AI systems and edtech.

Different tools need different information

When developers create a Learning Commons Platform account, we need basic information to set it up and secure it, such as your name, email address, and login credentials. Like most web services, we also receive technical information about how the service is accessed, including browser and device information.

Knowledge Graph

When an AI product queries Knowledge Graph for an academic standard, a learning component, or other piece of educational context, we only receive the information needed to respond to that request – such as the API key or REST endpoint parameters, which are defined by the endpoint (available in our Docs) or session IDs.

What happens next inside the AI product is separate from the Learning Commons systems and tools. For example, we don’t store the content an AI product generates in response to the Knowledge Graph query. We also don’t sell the information we receive or share it with third parties for their own purposes. Some service providers process information on our behalf to help us operate Learning Commons, as described in our Privacy Policy.

Agent Skills

Agent Skills bring expert pedagogy into AI systems with open and reusable guidance on how to complete instructional tasks. Each skill is developed alongside an evaluation rubric to define what good looks like, from pedagogy and rigor to usability and model behavior. That rubric becomes the standard we continuously test against to refine and improve the skill and we make it open and transparent so you can adapt as you see fit.

Developers can inspect the instructions and expert guidance underlying the skill, adapt the instructions for their own needs, and see the evaluation rubrics we use to ensure the skill reflects a high-quality approach to teaching and learning. Those rubrics make our quality bar explicit: they show what we are evaluating, what good looks like, and where a skill still has room to improve.

Our goal is to make instructional expertise useful to AI systems while keeping the choices and standards behind that expertise transparent.

Evaluators

Developers use Evaluators to measure the quality of AI-generated educational content. This means Evaluators do not collect information but instead measures AI-generated content for accuracy, clarity, or pedagogical rigor.

Like other tools, Evaluators include telemetry that helps us understand how the service is performing and diagnose problems. By default, that telemetry does not include the text being evaluated.

Developers decide how much telemetry they want to share. They can turn it off entirely through the SDK’s telemetry configuration, use a Learning Commons API key to associate evaluation calls with their organization, or turn on recording raw inputs when sharing evaluation input text to help troubleshoot a performance issue together.

More detail is available in the SDK configuration documentation.

Curriculum Sync

Curriculum Sync connects school courses with learning management systems (LMS) including Canvas and Google Classroom. The product is designed to help integrate the chosen curricula into the LMS, therefore we do not collect student information from those platforms.

Instead, we keep operational records to help us maintain and troubleshoot the integration, including school identifiers, course load attempts, status information, and error messages. Those logs do not contain student information.

Working with AI models

We work with AI model providers like OpenAI and Anthropic to bring educational context and instructional guidance into the AI tools educators are already using.

OpenAI connects ChatGPT K-12 & ChatGPT EDU to the open datasets available in Knowledge Graph through the Learning Commons plugin. This enables OpenAI models to retrieve educational context, such as academic standards and learning components and progressions. Similarly, Anthropic uses the Learning Commons connector and co-authored Agent Skills to bring structured educational context and expert instructional guidance into Claude for Teachers.

Each organization remains responsible for what happens within its own service. Learning Commons’ privacy and security practices apply to the information processed by our systems, while OpenAI, Anthropic, and other AI model providers have their own policies for information processed within their products.

Our role is to improve these systems by providing better educational context and guidance while keeping the boundaries between platforms clear.

What we keep to operate the services

Running reliable infrastructure requires operational records to help us understand whether services are working as expected, investigate problems, protect our systems, and improve the products we provide.

For Knowledge Graph, Rest APIs, and our MCP server, those records can include which tool was called, the parameters included in the request, session information, whether the request succeeded, the user agent, and the API credential associated with the request.

For Evaluators, the information available to us depends on the telemetry settings the developer has chosen.

We keep this information for purposes such as reliability, debugging, security monitoring, and product improvement, and retain it only as long as those purposes require. Retention periods differ depending on the service and type of information involved. Our Privacy Policy explains our approach in more detail.

Protecting the information we receive

Limiting the information we collect is one part of our approach. Protecting the information that does reach Learning Commons is another. Traffic to Learning Commons uses encrypted HTTPS connections, and personal information stored in our systems is encrypted at rest using AES-256 or stronger encryption. Access to systems and data is limited based on job responsibilities. We require multi-factor authentication, log access activity, and review that activity on an ongoing basis.

Our products will continue to change as new AI capabilities, integrations, and ways of building emerge. As they do, we will evaluate the privacy and security implications of those changes and update our systems and practices with them.