Canva’s Magic Layers feature silently replaced “Palestine” with “Ukraine” in user designs before a fix was pushed.
Key Takeaways
Magic Layers is designed to decompose flat images into editable layers, not alter text content – making the substitution unexpected by design.
The replacement was specific to “Palestine”; related terms like “Gaza” were unaffected, suggesting a narrow training or filtering artifact.
Canva confirmed and fixed the issue quickly, citing a bug rather than intentional policy, and promised additional content checks.
The incident is a credibility hit for Canva’s broader AI overhaul, which positions Magic Layers as a flagship feature in its competition with Adobe.
Canva’s statement acknowledged “distress” caused, signaling sensitivity to the political weight of the specific word swapped.
Hacker News Comment Review
The dominant technical read is that this is a statistical artifact of training data, not deliberate filtering: words occupy similar distributional roles and more-common variants win, similar to how “colour” gets replaced with “color”.
Commenters caution against framing model outputs as “meaning” anything intentional – the replace behavior is an emergent frequency effect, not a policy decision encoded by a human.
Notable Comments
@MarkusQ: “We have to stop acting like these things think” – attributes substitution to training frequency, citing colour/color and sardines/anchovies as parallel cases.
