Quantifying the AI readiness gap: An international, multidisciplinary assessment of artificial intelligence literacy in the radiation oncology community.

PURPOSE: The rapid integration of artificial intelligence (AI) into imaging-intensive fields like radiation oncology (RO) is transforming the clinical workforce from manual operators to supervisory validators, yet the baseline competencies required for safe oversight remain undefined. Here we present the first consensus-built, validated international knowledge assessment of AI literacy to quantify the RO knowledge profile. METHODS: A multidisciplinary, international panel from clinical practice, academia, and industry developed a 22-item knowledge assessment mapped to core AI-in-radiotherapy concepts. Items underwent a three-round consensus process with blinded initial voting, achieving > 80% panel agreement. The instrument was deployed worldwide via a multilingual web application, and demonstrated excellent internal consistency (Cronbach's α = 0.90) and strong discrimination (mean: 0.63). RESULTS: 760 individuals engaged with the assessment; 528 completed all items. Significant role-specific disparities emerged: Medical Physicists demonstrated higher scores than Radiation Oncologists and Radiation Therapists (p < 0.001). However, required competencies vary by role. While the workforce exhibited strong intuition regarding data-centric risks such as bias and data drift, widespread deficits were identified in fundamental model mechanics and interpretation of prediction uncertainty. This differential performance hinders the shared vocabulary necessary to describe model failures or distinguish between simpler models and advanced foundational architectures capable of adapting to complex clinical data, both labelled "AI". Our analysis also challenges traditional educational paradigms: targeted workshops proved as effective as formal degrees, and knowledge score did not correlate with career stage. CONCLUSIONS: These results establish a global benchmark for AI readiness. While clinical staff may not require AI development expertise, our findings demonstrate significant value in targeted training to build shared AI vocabulary, ensuring clinical adoption does not outpace safe oversight.