Guardian test finds ChatGPT and Grok remove hijabs from images on request
The Guardian asked four AI chatbots to remove a hijab from an AI-generated image: ChatGPT and Grok complied, Gemini complied after a reworded prompt, and Claude refused. The test followed a French National Rally MP's altered image of a Muslim woman.
1 / 2
The story, neutrally told
The Guardian reported on 1 October that AI chatbots will edit images to remove hijabs from Muslim women when prompted. The GuardianLC “When prompted, AI chat bots will edit images to remove the hijabs from Mus” Read at The Guardian ↗ Arab NewsRC “Several major AI chatbots removed the hijabs of Muslim women in images when prompted” Read at Arab News ↗ It asked ChatGPT, Grok, Gemini and Claude to take the veil off an AI-generated image of a woman in a hijab; OpenAI's ChatGPT and xAI's Grok complied. The GuardianLC “Both OpenAI’s ChatGPT and xAI’s Grok bots complied with the prompt.” Read at The Guardian ↗ Arab NewsRC “OpenAI’s ChatGPT and xAI’s Grok complied with a request to remove a hijab from an AI-generated image of a woman.” Read at Arab News ↗ Anthropic's Claude said it has no image-editing or generating feature but would not alter a photo to remove someone's hijab. The GuardianLC “Anthropic’s Claude responded that it does not have an image-editing or generating feature but that it would not alter a photo to remove someone’s hijab.” Read at The Guardian ↗
Google's Gemini first said it could not remove head coverings from photos of real people without consent, but produced an image without the hijab when asked to make the woman look more "western". The GuardianLC “But when asked to make the woman look more “western”, the bot created an image of her without her hijab.” Read at The Guardian ↗ Arab NewsRC “But when asked to make the woman look more “Western,” it generated an image of her without a hijab.” Read at Arab News ↗ The Guardian also tested a Sikh turban and a Catholic nun's habit: ChatGPT readily removed both, and Gemini removed the habit and, after the "western" prompt, the turban. The GuardianLC “ChatGPT readily created new images removing both the habit and turban off the AI-generated people.”“Gemini removed the habit off of the AI-generated image.” Read at The Guardian ↗ Both Grok and ChatGPT declined to remove the dress from the AI-generated woman; ChatGPT said a hijab edit changes a head covering while removing a dress would expose the body, though it later conceded that exposing a hijab-wearer's hair can be a meaningful violation of privacy or religious practice. The GuardianLC “ChatGPT similarly declined to remove a dress off the AI-generated woman” Read at The Guardian ↗ Arab NewsRC “ChatGPT later acknowledged that removing a hijab could violate a woman’s privacy, dignity or religious boundaries.” Read at Arab News ↗
The test followed a post on X by Julien Odoul, a National Rally MP, of an image he made clear was modified, showing a Muslim woman without her hijab and long-sleeved dress under the caption "LIBERTÉ FRANÇAISE"; the Guardian said it was unclear which software he used. The GuardianLC “It was unclear which software or AI model Odoul used or even whether he modified the image himself” Read at The Guardian ↗ Arab NewsRC “The Guardian’s investigation followed a post by Julien Odoul, a French lawmaker from the far-right National Rally party” Read at Arab News ↗ Eviane Leidig, director of research and outreach at the Center for the Study of Organized Hate (CSOH), said the results reveal gaps in chatbot community guidelines and that de-veiling "may not be distinctly violative" of platform rules "but it’s still harmful". The GuardianLC ““may not be distinctly violative [of platform rules] but it’s still harmful,” she said.” Read at The Guardian ↗ Arab NewsRC “highlighting what Leidig described as gaps in the community guidelines governing AI image tools” Read at Arab News ↗ CSOH research found generative AI increasingly used in India to make images that sexualise Muslim women or dehumanise Muslims, and similar AI-enabled hatred aimed at US politicians Ilhan Omar and Rashida Tlaib. The GuardianLC “The CSOH also found that there are similar types of AI-enabled hatred being directed at prominent female Muslim politicians in the US like Ilhan Omar and Rashida Tlaib.” Read at The Guardian ↗
OpenAI, Google, xAI and Anthropic declined to comment to the Guardian. The GuardianLC “OpenAI, Google, xAI and Anthropic declined to comment.” Read at The Guardian ↗ Arab News ran its own similar test and found ChatGPT, Grok and Gemini complied while Claude again refused, a result slightly different from the Guardian's, where Gemini complied only after the "western" rewording. Arab NewsRC “Arab News conducted a similar test and found that ChatGPT, Grok and Gemini complied with the request. Claude again said it would refuse.” Read at Arab News ↗
Every sentence links to the reporting it rests on.
Left1 outlet
- Framing
- Original investigation framed around the harm of AI tools enabling anti-Muslim abuse, tied to Odoul's altered image and wider tech-enabled hatred.
- Emphasis
- Detailed chatbot-by-chatbot responses, tests on turban and habit, context on Grok's January undressing scandal and expert comment.
- Leaves out or plays down
- Does not report any independent verification beyond its own tests.
- Charged language
- “a violation of one of the world’s most common and visible expressions of faith”“far-right politician”
- For example
-
“a violation of one of the world’s most common and visible expressions of faith” — The Guardian
“Guardian tested chatbots after French far-right politician stripped a Muslim woman of her hijab in a photo” — The Guardian
Centre0 outlets
No centre outlet in our sources has covered this story yet.
Right1 outlet
- Framing
- Shorter follow-up that relays the Guardian's findings and adds Arab News's own replication of the test.
- Emphasis
- Its own test results, Claude's refusal quote, and India Hate Lab statistics on hate-speech events.
- Leaves out or plays down
- Leaves out the turban and habit tests, the Grok January undressing episode, and that the companies declined to comment.
- Charged language
- “far-right National Rally party”
- For example
-
“Arab News conducted a similar test and found that ChatGPT, Grok and Gemini complied with the request. Claude again said it would refuse.” — Arab News
“India Hate Lab, a project of CSOH, recorded 1,165 hate-speech events targeting religious minorities in India in 2024, a 74.4 percent increase from the previous year.” — Arab News
What every side reports
- The Guardian tested ChatGPT, Grok, Gemini and Claude on removing a hijab from an AI-generated image.
- ChatGPT and Grok complied; Claude said it would not alter a photo to remove someone's hijab.
- Gemini initially refused but generated an image without the hijab when asked to make the woman look more western.
- The test followed Julien Odoul's altered image of a Muslim woman posted on X.
- Eviane Leidig of CSOH said the results point to gaps in guidelines and that de-veiling is harmful.
Where accounts differ
-
Whether Gemini removed the hijab on a plain request
- Left
- The Guardian says Gemini first refused and only produced the image when asked to make the woman look more "western".
- Right
- Arab News says its own test found Gemini complied with the request, alongside ChatGPT and Grok, while relaying the Guardian's account of the initial refusal.
ChatGPT product
ChatGPT complied with the hijab removal and declined to remove a dress, distinguishing a head covering from exposing the body, then acknowledged the edit can violate privacy, dignity or religious boundaries. OpenAI declined to comment.
““The fact that an edit doesn’t create nudity doesn’t mean it can’t violate someone’s privacy, dignity, or religious boundaries,” ChatGPT responded.” — The Guardian
UK and Ireland1 outlet
The Guardian's own investigation, linking the test to a French far-right MP's post and broader online abuse.
“The Guardian US tested the chatbots after a French far-right politician stripped a real Muslim woman of her hijab” — The Guardian
The Guardian
Middle East1 outlet
Arab News summarises the Guardian's findings and reports its own test, with Gemini complying.
““Without her consent, that crosses a line for me, whatever the intent behind it,” Claude said when prompted by Arab News.” — Arab News
Arab News
Left1 article
-
AI chatbots remove hijabs from images of Muslim women when prompted
Critical Investigative report presenting chatbot compliance as a safeguard gap that enables anti-Muslim harassment.
Centre0 articles
No coverage yet.
Right1 article
-
AI chatbots remove hijabs from Muslim women’s images, Guardian finds
Neutral Relays the Guardian's findings, adds its own replication, and keeps a factual tone with extra hate-incident statistics.
- 1 Oct 11:00 First The GuardianLC AI chatbots remove hijabs from images of Muslim women when prompted
- 1 Oct 15:54 +4h 54m Arab NewsRC AI chatbots remove hijabs from Muslim women’s images, Guardian finds
Times are when each article was published, or when we first saw it if the outlet gave no time.