Open-Sourcing Highly Capable Foundation Models: An evaluation of risks, benefits, and alternative methods for pursuing open-source objectives

TL;DR

The paper analyzes the risks and benefits of open-sourcing highly capable foundation models, arguing that while openness can enhance external evaluation, accelerate progress, and distribute influence, frontier models may present extreme risks that outweigh these benefits. It defines core terms, distinguishes OSS from AI release, and highlights two main risk channels: malicious use and the proliferation of unresolved flaws, with irreversible consequences once released. The authors propose a risk-based decision framework, advocate for safer alternative sharing approaches (e.g., staged releases, gated access, and audits), and offer governance recommendations for developers, standard-setting bodies, and governments. The work emphasizes that open sharing remains valuable but must be pursued with careful risk assessment, collaboration, and policy mechanisms to preserve benefits while mitigating potential harms.

Abstract

Recent decisions by leading AI labs to either open-source their models or to restrict access to their models has sparked debate about whether, and how, increasingly capable AI models should be shared. Open-sourcing in AI typically refers to making model architecture and weights freely and publicly accessible for anyone to modify, study, build on, and use. This offers advantages such as enabling external oversight, accelerating progress, and decentralizing control over AI development and use. However, it also presents a growing potential for misuse and unintended consequences. This paper offers an examination of the risks and benefits of open-sourcing highly capable foundation models. While open-sourcing has historically provided substantial net benefits for most software and AI development processes, we argue that for some highly capable foundation models likely to be developed in the near future, open-sourcing may pose sufficiently extreme risks to outweigh the benefits. In such a case, highly capable foundation models should not be open-sourced, at least not initially. Alternative strategies, including non-open-source model sharing options, are explored. The paper concludes with recommendations for developers, standard-setting bodies, and governments for establishing safe and responsible model sharing practices and preserving open-source benefits where safe.