Principles without provisions? Unpacking assessment in higher education AI policies

This study examines how university AI policies across Germany, Austria, and Switzerland address artificial intelligence in assessment and align with responsible AI principles. A census-based policy analysis was conducted on 73 institution-wide AI policy units from 95 German-speaking universities. Policies were coded along structural and regulatory characteristics, permitted AI use levels using an adapted AI Assessment Scale, and five responsible AI dimensions (Transparency, Fairness, Accountability, Academic Integrity, Human Oversight), complemented by a thematic analysis of responsible AI provisions. Policies are predominantly abstract (60.3%); although 47.9% are regulative in character, binding form rarely translates into concrete provisions. Of 70 assessment-relevant policies, 42.9% delegate AI use decisions below institutional level. Responsible AI principles are broadly invoked but unevenly operationalised: Transparency and Human Oversight most consistently, while Fairness and Accountability remain largely declaratory. Assessment orientation is predominantly summative, with Assessment as Learning nearly absent (7.1%).The findings reveal a systematic gap between normative positioning and operational governance. Future policies should translate principles into concrete frameworks that regulate AI while leveraging it for learning-oriented assessment design. Three levers are identified: defined disclosure formats, institutional minimum standards for delegated decisions, and institutionally licensed AI tools.