AI Transparency Statement

Socrates Educational Platform Effective Date: March 9, 2026 Last Updated: March 9, 2026

1. Our Commitment to Transparency

Artificial intelligence is central to how Socrates works. We believe you have the right to understand how AI is used on this Platform, what data it receives, how its outputs are validated, and what its limitations are.

This document explains all of that in plain language.

2. What AI Does on the Platform

2.1 Homework Analysis

When a student uploads homework (as a photo, document, or text), AI reads the content and identifies:

The subject area (math, science, history, etc.)
Specific concepts and topics covered
The difficulty level relative to the student's grade
Key learning objectives

This analysis is the foundation for everything else the Platform does with the homework.

2.2 Educational Game Generation

AI generates interactive educational games based on the homework analysis. There are six game types:

Quick Quiz: Multiple-choice and short-answer questions
Concept Blitz: Fast-paced concept matching
Myth Busters: True/false with explanation
Odyssey Trail: Narrative-driven learning journey
Socratic Showdown: Debate and argumentation
Labyrinth of Logic: Step-by-step problem solving

Games are generated as structured JSON data — not as executable code. The AI produces the content (questions, answers, explanations, narratives), and our pre-built React components render it. No AI-generated code ever runs in your browser.

2.3 Socratic Tutoring

The help system uses AI to guide students through concepts they find difficult. It follows the Socratic method:

Escalation Level 0-1: Guiding questions to help the student think through the problem themselves.
Escalation Level 2: Hints and partial explanations.
Escalation Level 3+: Direct explanations when the student is genuinely stuck.

The system adapts based on how the student responds, becoming more helpful as needed rather than immediately giving away answers.

2.4 Study Audio Generation

AI creates spoken study summaries based on uploaded homework, allowing students to review material by listening.

2.5 Note Evaluation

When students write study notes, AI can evaluate them for completeness and accuracy relative to the source material.

3. What AI Does NOT Do

It is equally important to understand the boundaries of AI on this Platform:

AI does not make educational decisions. It does not determine grade placement, recommend academic tracks, assess learning disabilities, or make any decisions that affect a student's educational path.
AI does not replace teachers. The Platform is a supplementary tool. It does not provide certified instruction, issue grades, or serve as a substitute for classroom learning.
AI does not grade students. Game scores reflect performance on AI-generated activities, not official academic grades.
AI does not discipline students. All moderation actions (warnings, suspensions, terminations) are taken by human administrators.
AI does not make decisions about accounts. Account creation, modification, and deletion are controlled by users and administrators.
AI does not communicate with students outside the Platform. AI does not send emails, messages, or notifications. It only generates content that appears within the Platform's interface.
AI does not access the internet. AI processes only the data we send it (homework content, grade level, subject context). It does not browse the web, access external databases, or pull in information beyond its training data and the content we provide.

4. AI Providers

We use two AI providers, and the Platform is configured to use one at a time:

4.1 Anthropic (Claude)

Model: Claude (via Anthropic API)
Capabilities used: Text analysis, image analysis (homework photos), text generation (games, tutoring, study audio)
API data policy: Data sent via the API is not used to train Anthropic's models. See Anthropic's privacy policy.

4.2 OpenAI (GPT-4o)

Model: GPT-4o (via OpenAI API)
Capabilities used: Text analysis, image analysis (homework photos), text generation (games, tutoring, study audio)
API data policy: Data sent via the API is not used to train OpenAI's models. See OpenAI's API data usage policy.

4.3 What Data AI Providers Receive

Data Sent	Data NOT Sent
Uploaded homework content (text extracted from images, document text, direct text input)	Student name
Images of homework (when vision analysis is needed)	Student email address
Student's grade level	Student's school name
Subject area context	Family information
Previous conversation context (for tutoring continuity within a session)	IP address or device information
	Other students' data

We intentionally exclude personally identifiable information from AI requests. AI providers receive the educational content needed to generate useful outputs, and nothing more.

5. How AI Outputs Are Validated

AI is not infallible. We implement multiple layers of validation to ensure AI outputs are safe and well-structured:

5.1 Schema Validation (Zod)

Every AI response must conform to a predefined schema before it reaches the student. For example, a game response must include specific fields (questions, answer options, correct answers, explanations) in the correct format. If the AI response does not match the schema, it is rejected and regenerated.

5.2 Content Proxy and Sanitization

Any content that references external resources passes through our server-side content proxy, where it is:

Checked against a domain allowlist (only trusted educational domains are permitted)
Sanitized using DOMPurify to remove any potentially harmful content
Served to the browser through our proxy, never directly from third-party sites

5.3 No Code Execution

AI generates data (JSON), not code. The Platform's frontend components interpret this structured data to render games, display tutoring responses, and play study audio. There is no mechanism for AI-generated code to execute in the browser.

5.4 Human Oversight

Administrators can review any AI-generated content through the admin dashboard.
Guardians can view their children's AI interactions, including tutoring conversations and generated game content.
Content flagged by automated systems or user reports is reviewed by human administrators.

6. Data Retention by AI Providers

Both Anthropic and OpenAI operate under API data policies that provide zero-retention or limited-retention guarantees:

Anthropic (API): Does not retain API inputs or outputs for model training. May retain data temporarily for abuse monitoring, per their policy.
OpenAI (API): Does not use API data to train models. May retain API data for up to 30 days for abuse monitoring, after which it is deleted, per their policy.

We select and configure AI providers to minimize data retention. We do not opt in to any data sharing or training programs.

7. Limitations and Accuracy

7.1 AI Can Be Wrong

AI-generated educational content may contain factual errors, incorrect answers, misleading explanations, or inappropriate difficulty levels. While our schema validation catches structural problems, it cannot guarantee factual accuracy.

7.2 AI Reflects Its Training Data

AI models are trained on large datasets and may reflect biases present in that data. We mitigate this through prompt engineering and output validation, but we cannot eliminate all bias.

7.3 AI Has Knowledge Cutoffs

AI models have training data cutoffs and may not reflect the most recent information in rapidly evolving fields.

7.4 AI Is Not a Substitute for Professional Guidance

The Platform should not be used as a substitute for professional educational assessment, special education services, or mental health support. If a student is struggling academically, consult a qualified educational professional.

7.5 Reporting Errors

If you encounter AI-generated content that appears incorrect, inappropriate, or harmful:

Students: Tell your guardian.
Guardians: Email [email protected] with details, including screenshots if possible.
Administrators will review flagged content and take corrective action.

8. Your Choices

Guardians can review all AI interactions their children have through the guardian dashboard.
Guardians can request that their child's data not be sent to AI providers. Note: this will disable the core educational features (game generation, tutoring, study audio) for that student's account.
Anyone can request information about how AI processed specific data by contacting [email protected].

9. Changes to AI Usage

If we make significant changes to how AI is used on the Platform — such as adding a new AI provider, expanding the types of data sent to AI, or adding new AI-powered features — we will:

Update this AI Transparency Statement.
Notify guardians by email at least 30 days before the change takes effect.
Obtain fresh parental consent where required by COPPA (e.g., if a new data category is shared with AI providers).

10. Contact

If you have questions about AI usage on the Platform:

Email: [email protected]