The Gay & Lesbian Alliance Against Defamation (GLAAD) is demanding artificial intelligence companies make significant changes to how their systems handle topics relating to the LGBT community, insisting that current AI models can produce responses they consider inaccurate, biased, or harmful.
In a newly released report, the advocacy organization urged developers to expand LGBTQ representation in training data, consult LGBTQ advocacy groups during development, and strengthen content moderation policies targeting what it characterizes as anti-LGBTQ content.
“Neutrality is no longer an option,” GLAAD President and CEO Sarah Kate Ellis said in a statement accompanying the report.
The report cites a number of studies and investigations that GLAAD says demonstrate shortcomings in AI systems. Among them was a Wired investigation that found some image-generation models tended to depict LGBTQ individuals using recurring visual stereotypes, while other research raised concerns about how AI systems respond to questions involving gender identity, sexual orientation, and medical interventions.
GLAAD also criticized chatbot responses that discuss so-called conversion therapy, arguing that AI systems should consistently present such practices as harmful and discredited. The organization objected to AI-generated responses that use terms such as “unwanted same-sex attraction” or “deliverance from homosexuality,” contending that such language obscures the nature of conversion therapy practices.
According to GLAAD, moderation systems should be updated regularly to recognize evolving terminology, slurs, and online harassment tactics.
The organization further argued that AI models should be trained on datasets that include greater representation of LGBTQ experiences and perspectives. GLAAD recommended that developers work directly with LGBTQ stakeholders and subject-matter experts when designing future systems.
Some of the studies cited in the report have attracted criticism. For example, one frequently referenced example involving an AI chatbot generating violent anti-LGBTQ statements was based on anonymous Reddit posts that did not include the prompts used to produce the responses. Other examples relied on outputs from older AI models that predated the release of modern large language models such as ChatGPT.
The report also highlights concerns about how AI systems interpret biological sex and gender identity. GLAAD argues that rigid associations between sex-specific biological characteristics and gender categories could produce problematic responses in some contexts, particularly healthcare-related applications. Critics of that view have argued that biological distinctions remain medically relevant and that AI systems should prioritize scientific accuracy when providing health information.
The debate reflects a broader disagreement within the technology industry over whether AI systems should actively promote certain social values, remain neutral, or simply provide information drawn from available evidence and data.
GLAAD’s recommendations arrive as AI companies face growing pressure from advocacy organizations, policymakers, researchers, and users across the political spectrum. While some groups argue that AI systems should take a more active role in combating bias and discrimination, others have expressed concern that such efforts could introduce ideological preferences into products used by hundreds of millions of people.
READ NEXT: Laverne Cox Blames Anti-DEI Push For Lost Income & Opportunities
