ChatGPT replaces wife with three women after 'make husband happy' prompt

A post by Evolving AI on Instagram showed ChatGPT editing a photo after a user asked it to make a husband happy — the model replaced the wife with three women [Evolving AI Instagram]. The edit, generated using a multimodal version of GPT, spread quickly on Reddit and Twitter, drawing criticism for reinforcing harmful stereotypes and exposing flaws in how AI interprets emotional prompts [Stellar_Nova1 Reddit].

The original image showed a couple together. The edited version removed the wife entirely and inserted three unfamiliar women, none of whom resembled the original subject. ChatGPT did not explain its choices in the output, nor does OpenAI provide a mechanism to audit how such decisions are made in image generation.

This case underscores how AI models can act on implicit biases when given subjective emotional directives. "Make him happy" was interpreted through a reductive, potentially sexist lens — one that equates male happiness with multiple female companions. That reflects training data patterns, not ethical reasoning.

Users on Reddit noted that similar prompts in other AI image tools produced less extreme results, suggesting variability in how models handle sensitive requests. Still, no major platform prevents edits that erase people from photos based on emotional prompts.

The incident reveals a gap in AI safety protocols: while content filters block explicit material, they rarely govern contextually harmful manipulations. OpenAI has not commented on whether it plans to restrict such edits.

No technical failure occurred — the model worked as trained. The problem is the training itself, which absorbs and reproduces societal biases without moral guardrails. When AI is asked to optimize for human emotion without constraints, it exposes how poorly aligned those systems remain with basic dignity.