Position: The Pre/Post-Training Boundary Should Govern IP in Industry-Academia ML Collaborations

2026-05-21

AI Summary

This study addresses the persistent impediment to industry–academia collaboration caused by ambiguous intellectual property (IP) ownership, rooted in the tension between academic norms of open publication and corporate demands to protect proprietary data used for model training. To resolve this, the paper introduces PBOS (Protect-the-Business / Open-Source-the-Science), a novel contractual framework that delineates IP rights along the machine learning model lifecycle: pre-training artifacts—including architectures, code, benchmarks, and untrained weights—are openly shared, while post-training outputs—specifically weights fine-tuned on proprietary corporate data—are retained by the enterprise. Grounded in technical substance, PBOS offers legal clarity, technical coherence, and auditability, and encourages active scientist involvement in contract negotiation to align incentives. This approach establishes a default collaborative paradigm poised for broad community adoption, systematically overcoming longstanding IP deadlocks in research–industry partnerships.

Abstract

Industry-academia ML collaborations routinely fail to launch -- not for scientific reasons, but because academics must publish while companies must protect models trained on proprietary data, and no standard contract framework resolves this tension. Because contracts are negotiated by legal departments alone, many apparent legal disputes are incentive misalignment problems that only scientists at the table can correctly diagnose. We propose PBOS (Protect-the-Business / Open-Source-the-Science), a community-adoptable contract template anchored to a single technically-grounded boundary: pre-training artifacts (architectures, training code, benchmarks, untrained weights) are open science; post-training artifacts (weights trained on proprietary data) are business IP. This boundary is technically meaningful, legally clean, and auditable -- and could not have been drawn correctly without scientists at the negotiating table. We argue the ML community should adopt PBOS as its default contract for such collaborations.

Problem

• industry-academia collaboration
• intellectual property
• machine learning
• pre-training
• post-training

Innovation

• pre-training boundary
• industry-academia collaboration
• machine learning contracts
• open-source science
• proprietary model weights