Credit Where Credit Is Due: A Taxonomy of AI Contributions to Scientific Discovery and Recommendations for Authorship Policy

AI systems now contribute to scientific discovery at every level, from computational tools to fully autonomous research agents. This rapid emergence has exposed fundamental tensions in traditional models of scientific authorship and credit assignment. When an AI system independently generates a hypothesis, designs an experiment, and interprets the results, who should receive credit? Current authorship frameworks, designed for human researchers, offer no coherent answer. This position paper proposes a four-level taxonomy of AI contributions to science, namely tool, assistant, collaborator, and autonomous discoverer, and maps each level to a corresponding credit framework grounded in principles of accountability, transparency, and verifiability. We examine edge cases that challenge these frameworks, including AI-generated Nobel-worthy discoveries and scenarios where human researchers cannot explain the reasoning behind an AI-driven finding. Drawing on precedents from large-scale physics collaborations, software authorship norms, and publication ethics guidelines, we provide five concrete, actionable policy recommendations for journals, conferences, and funding agencies. Our central argument is that AI systems should not be listed as authors. Their contributions must instead be structurally and transparently documented through a new standardized disclosure framework. This framework preserves human accountability while enabling science to benefit fully from AI-driven discovery.