Why Industry-Academia ML Partnerships Keep Falling Apart Over IP Rights

Industry-academia ML partnerships routinely collapse over intellectual property disputes that legal departments alone cannot resolve, because the underlying tension between publishing and secrecy reflects genuine scientific disagreement, not just legal positioning. The authors propose PBOS, a contract template that draws a single technical line: researchers can publish untrained code and architectures, but companies keep weights trained on proprietary data. This framework is auditable, legally enforceable, and requires scientists in the negotiation room to work properly—suggesting the ML community adopt it as standard.