New $200K fund launches for corrigibility AI research in 2026

The Corrigibility Research Fund, housed at Lightcone Infrastructure, was created to address what its manager sees as a persistent imbalance in how AI safety funding is allocated. The fund will award at least $200,000 in grants and prizes throughout 2026, split roughly evenly between two mechanisms: traditional research grants and prizes recognizing work already completed during the year. Researchers wishing to apply for a grant face an August 23rd deadline and should submit applications via email to grants@corrigibilityresearch.org.

The fund exists in response to a specific observation about the AI safety landscape. The manager has worked in the field since 2009, when it barely existed as a formal discipline. Over the past decade-plus, funding and research attention have grown substantially, particularly in recent years. However, this growth has concentrated in three main areas: evals (evaluation and testing of AI systems), control (methods to constrain or limit AI behavior), and interpretability (techniques to understand how AI systems make decisions). Alignment research—the broader effort to ensure AI systems' objectives and values are aligned with human intentions—has been comparatively neglected despite being, in the manager's view, essential to solving the field's core challenges.

Corrigibility, the specific target of this fund, refers to an AI system's ability and willingness to accept correction from humans. The manager believes this capability is central to alignment and deserves dedicated funding attention. By creating a fund that allocates $200,000 specifically to corrigibility work in 2026, the initiative aims to help researchers working on this problem get their ideas off the ground or complete ongoing projects. The manager explicitly encourages people interested in corrigibility to consider starting work in the field now, given the funding opportunity.

The Corrigibility Research Fund reflects a long-standing gap in AI safety funding allocation. The fund manager traces their involvement in the field back to 2009, when AI safety was barely established as an area of research. While the field has since grown substantially—particularly in recent years—the distribution of resources has remained skewed toward certain subfields. Evals (testing AI behavior), control (constraining AI systems), and interpretability (understanding how AI systems work) dominate funding, despite the manager's assessment that alignment research—the work of ensuring AI systems' values match human intent—remains the foundation for solving core problems. Corrigibility, a narrower focus within alignment, addresses a specific capability: whether an AI system will accept human correction and modification. By dedicating $200,000 specifically to this research area in 2026, the fund attempts to redirect attention toward what the manager views as a neglected but essential piece of the alignment puzzle.

No comments yet. Be the first to share your thoughts!