Build for Everyone: A Framework for LGBTQ Representation and Safety in Al

This report synthesizes findings from academic and industry research, civil society documentation, journalism, and GLAAD’s own monitoring to provide a comprehensive examination of AI’s impacts on LGBTQ people. Key findings include:

1. If AI fails LGBTQ people, it fails everyone. Companies must continue to work to improve AI equity, counter bias, and strengthen the safety and quality of AI products for LGBTQ people and for everyone.
2. AI must authentically represent the diversity of LGBTQ lives, stories, and experiences. If LGBTQ topics and issues aren’t accurately represented during foundation model development or in fine-tuning, AI systems tend to perpetuate biased or stereotypical assumptions.
3. AI companies must continually update training datasets and fine-tune models to reflect how hate and misinformation evolve online. To better protect marginalized communities, models must have regular context updates as new terms, language, and social and cultural tensions emerge.
4. AI developers and deployers must prioritize data protection and privacy. Sensitive data collection, profiling, and inadequate safeguards expose LGBTQ people to discrimination, surveillance, and potentially physical harm.
5. AI content moderation must be implemented responsibly—it should enforce hate and harassment policies, but not suppress LGBTQ voices. AI can play an important role in scaling content moderation, but when moderation systems lack nuance, transparency, and human oversight, they can both fail to curb harassment and wrongly suppress legitimate LGBTQ content.
6. There is an urgent need for thoughtful industry oversight to protect users, especially historically marginalized communities. From discrimination and surveillance to misinformation and data misuse, many of the broad risks associated with AI fall disproportionately on marginalized groups, pointing to the need for careful regulatory oversight in collaboration with civil society.