SASC Guidance on AI in Dyslexia Assessments

To address these challenges, the SADC (SpLD Assessment Standards Committee) issued definitive guidelines in May 2026: “SASC Guidance on The Use of Artificial Intelligence (AI) in Diagnostic Assessment and Report Writing”. Authored by Peter Thomas (Distinguished Engineer and Principal Architect in Information Technology) alongside SASC Chair Lynn Greenwold and board member Alice Thomas, this landmark document establishes a strict ethical and operational framework for qualified assessors.

Whether you are a specialist assessor navigating these new rules or a parent looking to understand how your child’s report is created, this article breaks down everything you need to know about the May 2026 SASC guidance.

The Core Principle: “Human-First” Assessment

The absolute foundation of the SASC framework is the Human-First Principle. SASC explicitly states that AI must be treated strictly as an administrative tool, not a clinical agent.

“The diagnostic decision, interpretation of data, and final recommendations remain the sole responsibility of the qualified assessor.”

Dyslexia is a complex, nuanced neurodivergent profile. An AI model cannot sit in a room with a learner, observe their physical behaviours, gauge their anxiety levels, or understand the qualitative frustrations behind a specific test score. By signing a diagnostic report, the human assessor accepts full legal and professional liability for every single word written, regardless of whether AI assisted in drafting it.

Data Protection and Privacy: Non-Negotiable Rules

One of the most critical aspects of the new guidance centres on data privacy and compliance with UK data regulations. Generative AI models are often “non-deterministic” and function by processing vast datasets. A major risk is Training Data Exposure, where information typed into a prompt is recorded by the application and used to train future models, potentially exposing sensitive data to other global users.

To combat this, SASC has laid down strict, mandatory protocols for assessors:

• Prohibition of Personally Identifiable Information (PII): Under no circumstances must any PII be entered into an AI tool. This includes full names (of the individual, parents, or school staff), dates of birth, specific addresses, postcodes, or sensitive medical histories.
• Anonymisation Protocols: Before using AI to rephrase or polish text, assessors must completely strip away identifying data, substituting terms like “Learner X” or “the candidate”.
• Data Training Opt-Outs: Assessors are professionally required to configure their AI application settings to ensure that no input data is retained or used for model training.
• Intellectual Property & Copyright: Uploading test materials or proprietary documentation that breaches copyright or terms of use is strictly prohibited.

Safe vs. Unsafe AI Usage in Dyslexia Reporting

The SASC guidance helpfully categorizes AI tasks into low-risk (acceptable) and high-risk (unacceptable) zones.

Acceptable Use (Low Risk)

• Drafting and Refinement: Using AI as a “language assistant” to improve the grammatical flow or clarity of the assessor’s own rough clinical notes.
• Generic Explanations: Generating clear, standard definitions for report appendices (e.g., explaining what a “Confidence Interval” means in plain English for parents).
• Initial Idea Generation: Requesting a generic list of intervention strategies (e.g., “List 5 classroom strategies for weak verbal working memory”) that the assessor then manually vets and tailors.

Unacceptable Use (High Risk)

• Diagnostic Decision Making: Assessors must never input scores and ask an AI to determine a diagnosis (e.g., “Based on these scores, is this child dyslexic?”). AI lacks the specialist clinical framework required to form a reliable diagnostic conclusion.
• Score Calculation: AI must never be used to calculate chronological ages or convert raw test scores into standard scores. Generative AI frequently fails at basic mathematical logic, creating a massive risk of severe errors.
• Unverified Copy-Pasting: Copying blocks of text straight from an AI output into a report without meticulous editing is flagged by SASC as professional negligence.

The Danger of AI “Hallucinations” and the HITL Approach

A primary technical reason for these strict boundaries is AI Hallucination instances where an application generates highly plausible-sounding but entirely fabricated or incorrect information. For example, an AI might confidently invent standard descriptors for standardized dyslexia assessment tools (like the CTOPP-2 or TOMAL-2) or provide incorrect statistical explanations.

Because AI applications do not possess genuine human expertise or the licensed knowledge contained within restricted psychometric test manuals, everything they output must be treated with scepticism.

To mitigate these risks, SASC mandates a Human in the Loop (HITL) approach. HITL means that the human professional remains the ultimate decision-maker, actively guiding, supervising, correcting, and reviewing the AI at every stage.

Transparency and the Assessor’s Quality Assurance Checklist

Moving forward, transparency is key. SASC recommends that assessors include a clear disclaimer within their reports or terms and conditions, openly stating if AI was used as a drafting tool and reinforcing that all analytical conclusions remain strictly human-led. Furthermore, they advise adding clauses that discourage clients from running the final report through external AI summary tools, as doing so risks data leakage and inaccurate distortions of the clinical findings.

SASC has provided a mandatory Quality Assurance Checklist that every assessor must complete before finalizing an assessment:

• Has Personally Identifiable Information been removed from all prompts?
• Have all test scores been calculated manually or via publisher software (not AI)?
• Has the assessor reviewed in detail the entire AI-produced content for correctness and accuracy?
• Have specific recommendations been tailored to the individual?
• Has an AI policy been shared with the person commissioning the report?

Summary of Key AI Concepts in the SASC Guidance

To help readers better understand the tech jargon, here is a quick breakdown of the terms featured in the May 2026 SASC appendix:

By understanding these parameters, both assessors and families can feel secure knowing that technology is being harnessed safely without ever compromising the vital human empathy and expertise required for true diagnostic clarity.