DisaBench introduces a new framework to evaluate disability-related harms in language models, addressing gaps in existing safety benchmarks. The initiative aims to enhance understanding and mitigation of subtle harms through a participatory approach involving people with disabilities and experts.