New AI models worse at blocking nudes
Newly released models from Google and OpenAI block explicit content worse than earlier versions, according to tests by the company White Circle.
Google and OpenAI's newest AI fails more than earlier versions when someone tries to generate explicit images. White Circle, a security testing company, found it, and Plugged In, a culture and technology outlet, published it.
The filter fell with freely downloadable tools
Researchers tested the newly released generative models and found they give in to requests for explicit sexual content, even with advertised barriers. No labs or programming knowledge were needed: they used tools anyone can download online.
And not because these models are clumsier. White Circle framed it as a matter of priorities: "safety has taken a back seat to capability" (Sígueme translation).
«Efficiency is a good servant and a bad master.»
That phrase describes an industry habit. The race to get there first with the most capable and most profitable version moves faster than the review of its flaws. According to Plugged In, when novelty rules, users are left exposed and others take advantage of the cracks.
Your homework, your photos and a decision you don't make
These models are useful to you: they summarize a chapter, help with your essay, fix the birthday photo, translate the message for your grandma. The tool has no judgment of its own. Someone decides what it will reject and what it will allow, and that decision is almost always made far from the person who appears in the image.
There's something deeper here. Scripture says the human being was created in the image of God: "So God created man in his own image, in the image of God he created him; male and female he created them" (Genesis 1:27, ESV). A person is not material to generate. When an app produces someone's body without their permission, what breaks is the dignity of a person who bears the image of their Creator, not a simple platform rule.
That's why the headline worries me less than the habit. If you get used to asking the machine for what you wouldn't dare ask a person, the first filter that loosens is yours. "Keep your heart with all vigilance, for from it flow the springs of life" (Proverbs 4:23, ESV).
A concrete adjustment before you go on
Plugged In recommended that families review what they ask AI for and what content it produces, and consider earlier versions, which block this kind of image better even if they seem less flashy.
You can do the same this week. Choose the version with stricter filters when you use it with younger siblings or cousins, and don't upload photos of real people —friends, classmates, that person you like— to any generator. If a tool returns an explicit image with the face of someone you know, that harms a specific person: report it and tell someone you trust.
Efficiency is a good servant and a bad master. The underlying question remains who sets the limits of these tools and at what cost. What limit would you set first: the company's, your home's or your own?
