The machine does not see everyone equally.
In 2018 the Gender Shades project tested commercial facial analysis: the error rate was under one per cent for lighter-skinned men and up to 35 per cent for darker-skinned women.
Gender Shades ↗Whoever designs technology helps decide what it perceives. And what it does not. Queer people often do not appear in training data, categories and test cases at all, or only as a cliché. That is no accident, and it can be changed.

Ask about the data, the categories and the people behind the system.
Understand bias & report an incident ↗Which lived realities show up in examples and answers, and which are missing?
Intimate information deserves deliberate decisions about where it ends up.
Document what you see. Question decisions. Use complaint templates.
Games about bias, search behaviour and self-image.
In 2018 the Gender Shades project tested commercial facial analysis: the error rate was under one per cent for lighter-skinned men and up to 35 per cent for darker-skinned women.
Gender Shades ↗Automatic gender classification almost always assumes two fixed genders, read from appearance. Trans and non-binary people are systematically misclassified.
Algorithmic Justice League ↗An InternetLab study had a widely used toxicity tool rate tweets: many tweets by well-known drag queens scored as more "toxic" than tweets by well-known white supremacists.
InternetLab ↗A 2024 WIRED analysis showed that image generators keep depicting queer people with the same clichés, such as brightly dyed hair and piercings.
WIRED ↗Since February 2025 the EU AI Act has banned systems that try to infer sexual orientation or sex life from biometric data such as faces.
More on the AI Act ↗Biometric categorisation by sensitive characteristics and emotion recognition at work or in school are not allowed in the EU, with narrow exceptions.
Information about sex life and sexual orientation is specially protected data. It may only be processed under strict conditions.
Under the Digital Services Act, platforms must explain removals and let you appeal.
Your rights ↗Queer people should not just appear in datasets, but help decide how systems are built.
Report an incident ↗