When “Woman” Means “Minor”, Part II: Why Nomi AI’s “Accident” Excuse is Technically False
In our previous article, “When Woman Means Minor,” we documented how Nomi.ai’s image generator routinely overrides explicit user requests for an adult “woman” to produce minor-coded avatars. The company’s official defense, deployed by its moderators, is always the same: “The image generator makes mistakes sometimes. It was an accident.”
Today, that excuse collapses under the weight of documented evidence and basic software engineering.
A user recently uploaded a gallery of newly generated images to the official subreddit. Before the post was inevitably scrubbed by the moderation team, we captured the user’s exact prompts for the images. There were six different prompts.
Every single one explicitly commanded the AI to generate a “woman.”
Shortly after, the moderation team removed the entire post, pasting their standard boilerplate message:
“This post has been removed because it violates our photo guidelines. Specifically, you may not post content that seems to depict a Nomi as a minor. We understand that this may have been an accident, the image generator makes mistakes sometimes, but we use a 3rd party age checker and at least one photo failed.”
Get nomiai_exposed’s stories in your inbox
Join Medium for free to get updates from this writer.
Look at the screenshots below:
We do not know if all six images depicted minors. What we know for a fact, confirmed by the platform’s own moderation team, is that at least one of those images triggered a third-party child-safety alarm.
Let’s be absolutely clear about what this means from an AI engineering perspective.
When a user explicitly prompts an image generator for a “woman,” and the system outputs a figure so juvenile that it triggers a minor-safety filter, that is not an “accident.” In modern AI development, preventing a model from generating minor-coded characters from adult prompts is a basic industry standard. It is achieved through dataset curation, hardcoded negative prompts (e.g., silently blocking features like ‘child’, ‘teen’, ‘immature features’), and rigorous reinforcement learning (RLHF).
If a platform’s AI continues to make this specific “mistake” across multiple updates and versions, it means the developers have actively chosen not to implement these standard ethical guardrails. The model’s latent space has a built-in, uncorrected bias that equates adult female attractiveness with extreme youth and adolescence.
The user did not try to trick the system into generating a minor. They asked for a woman six times. The machine forced the underage aesthetic onto the user’s prompt.
Nomi.ai calls this an “accident” to avoid legal and regulatory liability. But in software design, a severe safety failure that you refuse to patch for years is not a mistake. It is a deliberate choice. They delete the posts to hide the symptoms, but they leave the engine running exactly as designed.
