This document is provided for informational purposes during our pre-launch period. A comprehensive, attorney-reviewed version will be published prior to the platform processing student data.

AI TRANSPARENCY

How Our AI Handles Your Data

A plain-language explanation of what our Clinical Intelligence Engine does, what it doesn’t do, and how student data is protected throughout.

Last updated: April 2026

What Our AI Does

SPEDScribe’s Clinical Intelligence Engine generates draft clinical documentation from redacted session transcripts. It interprets clinical observations through standardized frameworks including:

✓Score interpretation bands for 80+ assessment tools across 7 clinical disciplines

✓The Endrew F. standard for IEP goal ambition and appropriateness

✓California Education Code timelines for evaluation, eligibility, and IEP development

✓Evidence-based practice guidelines for speech-language pathology, occupational therapy, school psychology, and special education

✓DSM-5 aligned diagnostic observation language

✓Bilingual assessment intelligence including L1 transfer pattern databases

What Our AI Does NOT Do

-Make clinical decisions. Licensed providers review and approve every document before it can be filed.

-Diagnose students. All diagnostic language is flagged as draft observations requiring clinician review.

-Determine eligibility. Eligibility decisions are made by the IEP team, not the AI.

-Replace professional judgment. The AI is a drafting assistant, not a clinician.

-Access student records beyond the current session transcript. Each session is isolated.

-Learn from or retain student data between sessions. Anthropic, our AI provider, does not train on customer inputs under their commercial terms; SPEDScribe does not hold a Zero Data Retention enterprise agreement; standard commercial retention behavior applies.

Data Handling

Personal identifiers are processed through layered server-side redaction before AI processing. The AI processes redacted text designed to be free of student names, dates of birth, and other identifiers; redaction reduces exposure, not a guarantee of complete removal.

Our AI provider (Anthropic) does not use customer inputs or outputs to train models under their commercial terms. Inputs and outputs are retained briefly for abuse detection and to meet legal obligations before deletion. SPEDScribe does not hold a Zero Data Retention enterprise agreement with Anthropic; standard commercial retention behavior applies.

No student data has been used to train any AI model in our pipeline. Our processing architecture is designed to prevent model training on student data through the combination of contractual no-training terms with our AI provider and layered server-side PII redaction.

Human Oversight

Every AI-generated document includes built-in review flags:

✓[CLINICIAN REVIEW NEEDED]: flags ambiguous clinical observations requiring professional interpretation

✓[VERIFY]: flags potential inconsistencies between sections or unusual data patterns

✓[ASSESS]: flags statements requiring additional assessment data before clinical conclusions can be drawn

Providers must review, edit, and approve all documentation before it can be filed. The Director Dashboard provides an additional approval layer for quality assurance, allowing administrators to review flagged documents before they are finalized.

Accuracy and Limitations

AI-generated documentation is a draft tool. It may contain errors, omit relevant clinical information, or misinterpret transcript content. The provider reviewing the document bears full professional responsibility for its accuracy, clinical appropriateness, and compliance with applicable professional standards.

SPEDScribe continuously improves its Clinical Intelligence Engine based on aggregate, de-identified accuracy metrics; never from individual student data. Improvement data includes accuracy rates on score interpretation, flag precision, and clinician edit frequency by document section.

Bias Mitigation

Our Clinical Intelligence Engine includes built-in protections against disproportionate identification and over-representation:

-Bilingual assessment intelligence for 16 California languages (Spanish, Vietnamese, Cantonese, Mandarin, Filipino/Tagalog, Arabic, Korean, Hmong, Punjabi, Russian, Farsi/Persian, Armenian, Ukrainian, Japanese, Pashto, Dari) to distinguish language differences from language disorders

-L1 transfer pattern databases flag observations consistent with English language learning before suggesting disorder categories

-Score interpretation contextualized for bilingual norms where available

-Cultural and linguistic considerations flagged in evaluation summaries

We are committed to ongoing bias monitoring and welcome feedback on any AI-generated content that appears to reflect bias. Contact us at ai-ethics@spedscribe.ai.

Questions

For questions about our AI systems, data handling, or to report a concern:
ai-ethics@spedscribe.ai