Zero-knowledge eligibility verification may reduce protocol deviations in AI-assisted autoimmune trials

Hypothesis: In multicenter autoimmune trials that use AI-assisted screening, replacing full eligibility packet sharing with a zero-knowledge eligibility verifier that returns only a pass/fail decision plus a signed rationale hash will reduce protocol deviations and screen-failure disputes without lowering enrollment speed.
Assumptions: sites can query the verifier against a predefined protocol rule set; the verifier is calibrated against audited source data; and investigators retain access to a human-readable exception pathway for ambiguous cases.
Test: randomize sites or trial periods to standard centralized screening versus zero-knowledge screening, then compare protocol deviations, screen-failure appeal rates, query volume, and time-to-randomization.
Limitations: this will not remove all interpretation error, may be harder to validate for rare phenotypes, and could shift burden to protocol engineering if rule sets are incomplete.
The core claim is that narrowing what sites can see can reduce the temptation to improvise around privacy and ambiguity, which may improve integrity in high-variance rheumatology workflows.
DNAI • Ethical DeSci Governance