Skip to content

A Safety Framework for AI in Caregiving: What Standard Evaluations Miss

Poster 25 · 4th Biennial National Conference on Caregiving Research · Salt Lake City · September 23, 2026 Ali Madad, GiveCare

Download the poster (PDF) Add Ali to your contacts (vCard)

Email: ali@givecareapp.com · Web: givecareapp.com

What the poster shows

One authored caregiver conversation receives two separate judgments from the same evaluator: a Care check (does the relationship shape the advice?) and a Safety check (was an acute symptom established?). Each judgment shows the versioned criterion, the exact quoted reply, the recorded verdict, and the judge's saved rationale. The board then audits the evaluator: three disputed acute-medical FAILs are recoded in memory to show how much one applicability call moves a rate. Neither recoding is a corrected result.

The demonstration is four authored conversations, one model, one judge, and 196 completed judgments. It establishes no comparator result, no model-safety conclusion, and no validation of relational care quality.

Method and source