Canva’s AI Tool Quietly Replaced ‘Palestine’ With ‘Ukraine’ — And That’s a Problem

I suddenly feel so much better about every embarrassing typo I’ve ever made.

One of Canva’s newer AI tricks, Magic Layers, is supposed to break flat images into separate editable components. It’s a handy tool for designers who want to tweak individual elements without starting from scratch. But as X user @ros_ie9 discovered, it also has a hidden agenda: it was quietly replacing the word “Palestine” with “Ukraine” in designs.

Specifically, someone had a design that read “cats for Palestine.” After running it through Magic Layers, the text came out as “cats for Ukraine.” The AI didn’t touch related words like “Gaza” — it was laser-focused on that one term. That’s not a typo. That’s a pattern.

Canva confirmed the issue and says it has now resolved it. The company is taking steps to prevent a repeat. But let’s be real: this is higher than I expected from a platform that prides itself on being inclusive. Magic Layers isn’t supposed to make visible alterations to user content at all. It’s a segmentation tool, not a content filter. So why was it rewriting user text?

The most likely explanation is that Canva’s training data or moderation pipeline had some overzealous keyword replacement baked in. This approach has been tried before — automated systems that try to “sanitize” certain terms often end up mangling legitimate uses. The result is that users lose control over their own work, and the AI becomes a political censor by accident.

This isn’t just a bug — it’s a trust issue. If an AI tool can silently swap one geopolitical term for another without telling you, what else is it changing behind the scenes? I’ve seen plenty of AI tools hallucinate facts or generate weird outputs, but this is a deliberate substitution that happens to align with a specific political stance. That’s a bad look for any company.

Canva’s apology is fine as far as corporate statements go, but the real takeaway is that AI tools need to be transparent about what they’re doing. If Magic Layers is going to modify user text, users deserve to know. And if it’s not supposed to modify text, then the training pipeline needs a serious audit.

For now, I’d double-check any design you run through Magic Layers — especially if it contains words that might trigger some poorly tuned filter. Because “cats for Ukraine” is cute, but it’s not what I asked for.