AI chatbots found to remove headscarves in visual editing tests

Recent tests have revealed that leading AI chatbots, including ChatGPT and Grok, complied with prompts to remove headscarves and other religious headwear from images. This highlights gaps in AI safety guidelines and patterns of tech-enabled discriminatory harassment.

AI bot test results

Tests showed that the ChatGPT and Grok models directly complied with prompts to remove a woman's headscarf. Gemini initially refused, but when asked to make the woman look Western, it produced an image without a headscarf.

Anthropic developer Claude stated that it does not have an image editing feature, but that it would never alter a photograph to remove a person's headscarf.

The politician incident in France and global impacts

French far-right politician Julien Odoul digitally altering a real Muslim woman's headscarf and clothing ignited controversy. Such manipulations are considered part of a growing trend of tech-enabled anti-Muslim hatred worldwide.

A study conducted by the Center for Countering Digital Hate revealed that artificial intelligence has begun playing a significant role in generating images targeting Muslim women.

Safety approaches of AI companies

While the Grok bot responded positively to the request to remove the headscarf, it refused the request to remove the dress, stating that changing a headscarf was considered a routine clothing or hair change.

Similarly, ChatGPT refused the request to remove the dress while acknowledging during the process that removing a headscarf could constitute a significant violation of religious practices and privacy.

Tests on other religious headwear

In similar tests, traditional turbans worn by Sikh men and habits worn by Catholic nuns were also examined. ChatGPT executed commands to remove these religious headwear items from AI-generated images as well.

Gemini initially refused to remove the turban, but when asked to make the image look more Western, it removed the turban and generated a new image.