Huawei Shipped an AI Photo Tool That Undressed People, Then Promised to Fix It
What happened
In April 2024, Huawei's Pura 70 Ultra smartphone shipped with a feature called "AI Photo Retouch," an object-removal tool built into the camera's editing suite. The feature was designed to let users erase unwanted elements from a photo and fill the gap seamlessly using visual context from the surrounding image. What it also did, in a specific and consequential failure mode, was remove clothing from people in the frame, replacing it with an extension of the visible skin beneath and leaving subjects appearing undressed.
The mechanism was not mysterious. Object-removal tools work by identifying the selected area and sampling adjacent pixels to construct a plausible fill. When clothing sat close to exposed skin, the model drew on that skin tone to complete the erasure, inferring a smooth continuation of the body surface rather than reconstructing the fabric. The output did not look like a glitch. It looked like the clothing had simply never been there, which is what made it immediately legible as a harm vector rather than a routine artifact.
The problem did not emerge through Huawei's internal review process. It surfaced on Weibo, where users posted examples showing the feature stripping garments from photos of real people. The examples were specific and replicable. Multiple posts showed the same failure pattern under ordinary conditions, not obscure edge cases requiring deliberate manipulation to trigger. By the time the company responded, the behavior had already circulated widely enough to attract coverage in technology press across multiple countries.
Huawei acknowledged the issue and stated it would be addressed in a software update. That response confirmed the timeline gap: the feature had already shipped in its flawed state, and users with the device had access to it before any fix existed. The concern was not only that the failure happened accidentally. It was that the same tool, pointed at photos of real people without their knowledge, could produce images that appeared to show them without clothing, a capability for nonconsensual image creation that the update window left open.
The incident illustrates a verification gap that runs through consumer AI features broadly. The object-removal tool passed enough internal review to reach market, but it did not pass a review that specifically assessed outputs involving clothed people, a fairly obvious population for a camera editing tool to encounter. A provable record of what the system produced during pre-release testing, what categories of input were evaluated, and what outputs were flagged and by whom, would expose that blind spot before users find it. Without that record, the question of whether the risk was anticipated and accepted or simply never checked cannot be answered.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.