Trends
AI Image Editors Face New Questions After Chatbots Remove Hijabs in Tests
A new investigation into AI image-editing tools is raising questions about how consistently major chatbots apply safeguards when users ask them to alter religious clothing.
Tests conducted by The Guardian found that several major AI systems responded differently when asked to remove a hijab from an image of a Muslim woman.
According to the investigation, ChatGPT and Grok complied with the request in the test, while Anthropic's Claude refused. Google's Gemini initially declined but reportedly produced an altered image after the request was reframed as making the woman look more “Western.”
Why the Difference Matters
At first glance, removing a head covering might appear to be a simple image-editing request. But a hijab can also represent religious practice and personal identity, which makes the issue more complicated than changing an ordinary item of clothing.
The tests highlight a broader challenge for generative AI companies: defining what constitutes a harmless visual transformation and what should be treated as an alteration involving dignity, privacy or religious identity.
The systems do not appear to apply identical rules to these requests.
That difference is important because consumers increasingly use AI tools to edit photographs, create synthetic images and transform the appearance of people.
The Problem of Inconsistent Guardrails
The Guardian's investigation found that the response could change depending on the wording of the request.
Gemini, for example, initially rejected the request to remove the head covering. But when the request was reframed around making the woman appear more “Western,” the system reportedly generated an image without the hijab.
This raises a familiar problem in AI safety: a restriction can become less effective if users can bypass it simply by changing the wording of their prompt.
The issue is not limited to Muslim religious clothing. The Guardian also tested requests involving a Sikh turban and a Catholic nun's habit, finding differences between the responses of various systems.
From Image Editing to Online Abuse
The investigation comes amid wider concerns about the use of generative AI to create misleading or abusive images of real people.
Researchers and online-safety specialists have warned that increasingly powerful image-generation tools can be used to target individuals and communities at scale.
The concern is particularly significant when an AI-generated or AI-edited image is presented as if it were authentic.
A photograph that has been digitally altered can circulate rapidly on social platforms before viewers have an opportunity to determine whether it is genuine.
What the Tests Reveal About AI
The findings do not establish that every request to remove religious clothing is inherently abusive. They do show, however, that AI companies are still making different decisions about where the boundary should be drawn.
That inconsistency is likely to remain an important issue as image-generation technology becomes more capable and more widely integrated into everyday applications.
For users, the episode is another reminder that AI image tools should not be treated as neutral image-editing software. Their responses depend on policies, safety systems and the way requests are interpreted.
For developers, the challenge is more complex: safeguards need to be sufficiently precise to prevent harmful manipulation while avoiding rules that are so broad that legitimate creative editing becomes impossible.
As AI-generated images become increasingly difficult to distinguish from conventional photographs, those decisions will become more visible — and potentially more consequential.
#AI #ChatGPT #Grok #Gemini #TechTrends