AI Transparency & Remediation Disclaimers

AI-assisted remediation models

ADAFix employs machine learning, optical character recognition (OCR), and multimodal vision models to assist authors and accessibility teams in remediating digital PDF documents.

AI-assisted remediation suggestions (including generated PDF tags, reading order, and image descriptions) are starting points to accelerate compliance workflows. Automated checks alone do not establish full legal compliance. Always inspect and verify output before distributing documents to assistive technology users.

• Structure Detection: Models propose logical tag hierarchies (Headings 1 through 6, Paragraphs, Lists, and Figures) based on visual layout cues.
• Reading Order Synthesis: Algorithmic sequence parsers determine natural column flows and content stream ordering for assistive technology.
• Draft Alternative Text: Vision models analyze images, charts, and diagrams to draft suggested descriptions for human review and refinement.

Human oversight and verification requirement

Automated tools and AI heuristics cannot replace professional human judgment. An automated check passing a PDF/UA or WCAG rule test does not automatically mean the content is comprehensible or contextually accurate for an assistive tech reader.

AI-drafted alternative text may vary depending on document context. Review descriptions alongside charts, scientific diagrams, and complex figures to ensure domain terminology and pedagogical accuracy.

• Course instructors and authors must review generated image descriptions to verify that pedagogical nuances and domain terminology are preserved.
• Reading order should be checked in complex multi-column, sidebar, or callout layouts to confirm pedagogical flow.
• Data tables with multi-tier nested headers or irregular spans require human review before publishing.

No legal certification or compliance guarantee

ADAFix is a productivity and remediation workflow tool. Use of ADAFix does not constitute a legal certification, audit guarantee, or legal representation of full compliance under Title II of the Americans with Disabilities Act (ADA), Section 504/508 of the Rehabilitation Act, AODA, or WCAG 2.2 AA.

Institutions remain responsible for their digital accessibility compliance programs and must maintain institutional quality assurance processes.

Model training and data privacy

We do not use customer uploaded course documents, unpublished research, syllabi, or institutional files to train, fine-tune, or improve public third-party commercial AI models.

Document prompts and image patches sent to AI inference endpoints are transmitted ephemerally under contractual terms that forbid vendor retention for foundational training.

Audit reporting & provenance records

Every completed job on ADAFix generates a transparent Results Report. The report clearly distinguishes between automated heuristics applied, items modified by the user, and remaining unresolved items flagged for manual specialist intervention.