Open-source software (OSS) development is a culture and set way of working that involves the free and open sharing of software projects for further study, modification, and use. For thirty years OSS has proliferated alongside (and often inside) private software, encouraging cooperation and improving software quality through community review.

However, in the past couple of years, decades-old OSS tradition has been challenged by the development of increasingly capable AI. Concerns about the potential harms of malicious misuse have yielded heated debate about the prudence of open-sourcing AI with many arguing that some models pose too high of a risk to be made available for public download.

The open-source AI debate has been a high-stakes dialogue, yet despite what, at its start, often felt like a stark open-v.-closed, us-v.-them standoff, it has evolved productively into collaborative conversations.

This report reflects on areas of emerging consensus, persisting disagreement, and unanswered questions to start thinking about next steps for safely pursuing openness benefits.

This report does not pit risks against benefits. Instead it looks into what kinds of evidence should be collected in deciding when to release an AI model, and how we might productively work to reduce any risks of openness where they do exist.