Research question
When does open release of capable models create unacceptable risk?
Key findings
- Open weights are irreversible and enable safeguard removal, but also auditability and competition - Proposes staged release and alternatives
Open-Sourcing Highly Capable Foundation Models
GovAI's analysis of open-weight release risks that framed the open-source debate around safeguard removal.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.