HAIM: Human-AI Music Datasets for AI Music Production Tracking Benchmark

Researchers introduce HAIM, a new dataset and benchmark for detecting AI integration across music production workflows, moving beyond binary AI-or-human classification to track granular stages of AI intervention including hybrid and mastered content. The work exposes critical limitations in current AI detection systems as generative music platforms like Suno and Udio achieve human-quality output.

The emergence of production-grade generative music tools has created a fundamental challenge for AI detection frameworks: real-world music production is increasingly hybrid, blending human and machine workflows in ways that resist simple binary classification. This research addresses a genuine blind spot in the AI safety and content verification landscape. Current detectors fail to account for realistic production scenarios where AI generates a track that humans then master, or where humans produce content that AI subsequently refines—scenarios now commonplace in professional studios.

The HAIM dataset represents a methodological shift recognizing that AI's integration into creative workflows is fundamentally different from previous detection challenges. Unlike deepfake or synthetic media detection, music production involves legitimate, collaborative use of AI tools. The research indicates that state-of-the-art detectors exhibit systemic flaws, suggesting existing approaches are inadequate for real-world deployment. This has implications for content platforms, music licensing bodies, and rights management systems that increasingly need to identify not just whether AI was involved, but how and to what degree.

For the music industry and platform developers, this benchmark establishes new standards for transparency and verification. Rights attribution, creator compensation, and licensing negotiations all depend on understanding production provenance at a granular level. The work signals growing maturity in AI detection research—moving from academic binary problems toward industry-relevant complexity. Stakeholders managing user-generated content or music licensing will need detection systems capable of this nuanced tracking. The HAIM dataset provides a foundation for developing such tools, though significant engineering challenges remain in deployment.

• →Current AI detection systems fail to account for hybrid human-AI music production, the dominant real-world paradigm
• →HAIM dataset enables tracking specific stages of AI intervention rather than binary classification, setting new detection standards
• →Generative music platforms achieving human-quality output have outpaced detection and verification infrastructure
• →Music licensing and rights attribution systems require granular production provenance tracking, not simple AI-or-human categorization
• →State-of-the-art detectors reveal systemic flaws that necessitate foundational research advances in AI music tracking