AI Bias Is Putting LGBTQIA+ People at Risk

Earlier this month, at Axios’ AI+ NY Summit, GLAAD CEO Sarah Kate Ellis raised the concern that AI bias was putting LGBTQIA+ people at risk. During the Summit, she previewed GLAAD’s new report, Build for Everyone: A Framework for LGBTQ Representation and Safety in AI, which documents the scope of these harms and lays out concrete recommendations for the industry. The report points to a problem that Partnership on AI has worked to address for several years: building algorithmic fairness while maintaining data privacy. As the pace of AI adoption and innovation accelerates, LGBTQIA+ people, along with other marginalized communities, are increasingly exposed to the risks of algorithmic discrimination: the systematic distortion in data or a system’s development that causes unjust outcomes or harm. For LGBTQIA+ communities, the consequences are particularly acute. Addressing bias in AI systems typically requires collecting demographic data from users to assess how a system performs across different groups. But for communities already at risk of discrimination or harm, that data collection process can itself become a source of harm. This is the problem PAI’s Participatory & Inclusive Demographic Data Guidelines were built to address. Bias in Action The impacts of AI bias on LGBTQIA+ communities are already visible. GLAAD’s report noted that in a 2024 UNESCO study Meta’s Llama 2 model generated negative content about gay people in approximately 70% of instances, producing statements that characterized gay people as criminals, outcasts, and abnormal. These findings reflect what happens when training data encodes existing bias and prejudice at scale. As GLAAD’s report notes, when foundation models carry these distortions, the errors propagate across the downstream applications built on top of them. The problem extends beyond biased outputs. AI systems can infer LGBTQIA+ identity from behavioral patterns, search history, and social connections that users never explicitly disclosed. Human Rights Watch has documented surveillance technologies used by governments to identify and target LGBTQIA+ people in countries where same-sex relationships are criminalized. The risks are compounded by a deteriorating political environment. According to the Trans Legislation Tracker, 797 anti-LGBTQIA+ bills have been filed in the U.S. so far in 2026. At the same time, AI is becoming more embedded in the systems people use every day, and when those systems encode bias, or when the data they collect is used in ways people did not anticipate or consent to, LGBTQIA+ people bear a disproportionate share of the consequences. “. . . communities affected by algorithmic discrimination need to be part of shaping the standards meant to protect them.” Guidelines for Safer Systems Addressing algorithmic bias requires demographic data so that organizations can understand how a system performs across different groups to detect and correct discriminatory outcomes. But collecting sensitive demographic data from users, including data related to sexual orientation and gender identity, introduces its own risks: miscategorization that erases certain identities from the assessment altogether, repurposing of data beyond its original scope, and inadequate security that exposes sensitive information. The Participatory & Inclusive Demographic Data Guidelines were developed to help organizations navigate this challenge responsibly. The Guidelines provide AI developers, technology teams, and data practitioners with guidance on how to collect and use demographic data for fairness assessments in ways that protect, rather than further expose, the communities most at risk. Seven equity experts advised on the development of the Guidelines, among them specialists in LGBTQIA+ justice. Our approach to developing the Guidelines reflected our belief that communities affected by algorithmic discrimination need to be part of shaping the standards meant to protect them. Data collection can have a chilling effect on the communities it is meant to serve, reducing the range of viewpoints and information people are willing to share when they know, or suspect, they are being watched. When Grindr shared users’ HIV status data with analytics firms, the number of users disclosing their status on the app dropped dramatically. This shows how data collection intended to serve a community can, if mishandled, cause that community to withdraw from spaces meant to support them. GLAAD’s report, which cites PAI as a key resource for responsible AI development, reinforces this approach, calling for privacy-by-design principles across the AI lifecycle and for meaningful civil society engagement with subject matter experts from the earliest stages of development. What Is Still at Stake The GLAAD report adds to a growing body of evidence that the AI industry has not yet adequately addressed these problems. But documentation alone is not enough. What’s needed are concrete practices, standards for how demographic data is collected, how consent is structured, how data is secured, and how communities most at risk of harm are involved in defining what fairness means for their own lives. Even with guidelines, implementation across the industry is uneven, regulatory frameworks remain incomplete, and the pace of AI deployment continues to outrun the development of adequate safeguards. As AI becomes further embedded in healthcare, employment, education, and public services, the consequences of getting this wrong grow more serious. For LGBTQIA+ people navigating both algorithmic discrimination and an increasingly hostile political environment, the margin for error is narrow. For communities that have the most to lose when AI systems fail, or when the data those systems collect is turned against them, trust has to be built through consistent, demonstrable practice, not stated commitments. We will continue working with our multistakeholder community to develop and strengthen the standards that make safe, responsible, and trustworthy AI possible. In an upcoming issue of AI&, we sit down with GLAAD to discuss their new report and what it will take to move the industry from stated commitments to demonstrable practice. Sign up below to receive the next issue.