Implicit Policies toward Genai in Educational Assessment: Identifying Acceptance Profiles Using Fuzzy C-Means

The integration of generative artificial intelligence (GenAI) into educational assessment has intensified a persistent tension between academic integrity and fairness, on the one hand, and the potential cognitive support that AI tools can provide, on the other. Grounded in the Artificial Intelligence Assessment Scale (AIAS), this study operationalizes teachers’ acceptance of GenAI use in assessment through five graded permission levels: (1) total prohibition (“No AI”), (2) AI permitted only for brainstorming/structuring with no AI-generated content included in the final submission, (3) AI permitted for editing only (no new content generation) with the original version attached, (4) targeted AI use allowed with mandatory human commentary and citation of AI-generated content, and (5) fully integrated use (“AI copilot”) without specifying what is AI-generated. Using survey data from 393 university professors in Western Romania, we applied Fuzzy C-Means clustering to identify latent acceptance profiles while allowing probabilistic (fuzzy) membership. The optimal five-cluster solution (R² = 0.400; BIC-minimizing) revealed clearly differentiated patterns, including a copilot-friendly profile, conditional-permissive profiles that endorse bounded and transparent uses but reject copilot integration, and larger diffuse segments characterized by broad non-endorsement of intermediate permission rules. Overall, the findings indicate that teachers’ GenAI assessment stances are not unidimensional; they cluster into distinct governance profiles shaped by preferences for control, transparency, and attribution. These profiles have direct implications for institutional policy, disclosure standards, and assessment design strategies that seek to balance pedagogical innovation with credible evidence of student learning.