AI Transparency

How Paideon Uses AI

Full transparency on our AI grading technology — what it does, what data it receives, and how teachers remain in control of every grade.

Last updated: March 10, 2026

No Student PII
Student names, classmate names, phone numbers, and addresses are automatically stripped from text before reaching the AI. Image/PDF submissions use prompt-level PII protection.
No Training on Your Data
Student submissions are never used to train or improve AI models.
Teacher Has Final Say
Every AI grade requires teacher review before it becomes final.

1. What AI Does in Paideon

Paideon uses AI to assist teachers in grading student work. The AI:

  • Reads student submissions (essays, short answers, documents)
  • Evaluates the submission against each criterion in the teacher's rubric
  • Suggests a score for each criterion with a brief reasoning explanation
  • Generates constructive student feedback based on the rubric criteria

The AI does not make final grading decisions. All AI-generated grades are presented to the teacher as suggestions, marked “Needs Review,” and require explicit teacher approval before they become part of the student's record.

2. What Data the AI Receives

Sent to AI

  • The text content of the student's submission (or a scanned image of handwritten work)
  • The teacher's rubric (criteria names, point values, level descriptions)
  • The assignment title and description

NOT Sent to AI

  • Student names (including classmate names mentioned in essays)
  • Student ID numbers
  • Email addresses, phone numbers, or street addresses
  • Class names or school names
  • Teacher names or account information
  • Any data from other students or other classes

Before a text submission is sent to the AI, our system automatically strips student names (the student's own name and all classmates in the same batch), email addresses, phone numbers, street addresses, and student IDs. The AI receives a prompt that says “Grade the following student submission:” followed by the stripped content — nothing more.

Image & PDF Submissions

For image-based submissions (e.g., photos of handwritten essays) and PDF documents, automated PII stripping is not technically possible — the content is visual, not text. In these cases, the AI is explicitly instructed via the system prompt to ignore any student names, emails, or other personally identifiable information visible in headers, margins, or metadata. This is a prompt-level control. Text extracted from PDFs and documents for other features (e.g., similarity detection) does go through automated PII stripping.

3. AI Model & Provider

  • Provider: Anthropic (San Francisco, CA)
  • Model: Claude (Sonnet class)
  • Processing location: United States
  • Anthropic's data policy: Anthropic does not use API customer inputs to train their models. This is explicitly stated in their privacy policy.

4. Data Retention by the AI Provider

Per Anthropic's API terms, API inputs are not stored long-term and are not used for model training. Anthropic may retain API logs for up to 30 days for safety monitoring and abuse prevention, after which they are deleted.

Since no student PII is included in API requests, even this temporary retention poses no risk to student privacy.

5. Human Oversight

Paideon is designed with a “human-in-the-loop” model:

  • AI-generated grades are always marked as “Needs Review” by default
  • Teachers see the AI's suggested score, reasoning, and confidence for each rubric criterion
  • Teachers can adjust any score, edit the reasoning, modify the feedback, or override the AI entirely
  • No grade is finalized until the teacher explicitly clicks “Approve”
  • Teachers can flag submissions for manual grading without using AI at all

6. Using Paideon Without AI

If your school or district policy prohibits AI-assisted grading, Paideon can be used as a manual grading tool:

  • Upload student work and rubrics as normal
  • Use the rubric interface to manually score each criterion
  • Write feedback manually in the provided fields
  • Export grades to Canvas CSV as usual

The platform's value — rubric management, roster tracking, grade organization, CSV export — works independently of AI grading.

7. Bias & Fairness

Because the AI receives no student identifying information (no names, no demographics, no prior grades), it cannot exhibit identity-based bias. Every submission is evaluated solely on its content against the rubric.

That said, AI models can have inherent biases in language evaluation. This is why teacher review is mandatory — the teacher's professional judgment is the final authority on every grade.

8. Questions

For questions about our AI practices:

Last updated: March 10, 2026