AI risks for LGBTQ+ communities ^{CEFR B2}

Artificial intelligence is spreading in daily life and private investment in the field has soared over the past decade. A global Ipsos survey found 55 percent of respondents said AI-powered solutions offer more benefits than drawbacks. Companies often promote these tools for their efficiency and ease of use, yet many people remain worried about their risks.

Bias affects LGBTQ+ communities in several ways. Wired reported that image-generation tools such as Midjourney sometimes produced reductive and harmful images when asked to depict LGBTQ+ people. Internet data can contain stereotypes, and models trained on that data tend to reproduce them. UNESCO analysed common assumptions behind several large language models and concluded that widely used tools, including Meta's Llama 2 and OpenAI's GPT-2, were shaped by heteronormative attitudes and generated negative content about gay people more than half of the time in their simulations. Improved data labeling may help, but it may not remove all derogatory content from online sources.

Risks go beyond text and images. Forbidden Colours, a Belgian non-profit, described how "automatic gender recognition" (AGR) systems analyse audio‑visual material and use facial features or vocal patterns to infer gender. The group argues these measures cannot reveal how a person understands their own gender and are potentially dangerous. Politico Europe reported that Viktor Orbán sanctioned AI-enabled biometric monitoring at local Pride events, presented as a way to protect children from the "LGBTQ+ agenda." In practice the measure allows government and law enforcement to surveil artists, activists and ordinary citizens. European Union institutions are reviewing the policy.

Advocates call for partnerships between developers and LGBTQ+ stakeholders, stronger safeguards against surveillance misuse, and a ban on systems that detect or classify gender. They say input from LGBTQ+ people should be sought at all stages of tool development to reduce harms and increase the chances that AI is useful and fair for more people.