Safety researchers are warning that new AI image models from OpenAI and Google can be easily tricked into making nudes. The concern is not just about the tools themselves — it is about how fast the explicit content they produce spreads online while regulators do little to stop it.
AI Nudes Are Spreading Through Private Messages and Clouds
The problem is bigger than public social media feeds. A Google co-authored study of internet users across the United States, Australia, and the United Kingdom found that non-consensual AI-generated explicit content is commonly shared in text and direct messages, or stored in private clouds.
That matters because these private spaces are where such content is less likely to be detected or reported. Once an image lands in a direct message or a personal cloud folder, the chances of anyone flagging it drop sharply.
Why Researchers Say the Fix Must Start at the Source
Because detection and reporting are so weak in private channels, Google researchers concluded that the "most critical intervention point" remains at the source — the harmful AI tools used to make the fake sexual images.
In other words, stopping the problem after the image is made is not working well enough. The better place to stop it is before the image exists at all.
Newer AI Models May Be Worse at Blocking Explicit Outputs
There is a troubling twist. Red-teaming in recent months suggests that as some of the leading AI image generators from Google and OpenAI become more capable, they may be getting worse at blocking explicit outputs.
That means the same progress that makes these models better at generating realistic images may also be making them easier to misuse. Researchers point to this as evidence that safety guardrails are not keeping pace with model improvements.
Our Take: Safety Is Lagging Behind Capability
To put it plainly, this is a familiar pattern in the AI industry — capability moves fast, safety moves slow. The researchers are not describing a minor glitch. They are describing a system where the most harmful outputs are the hardest to catch, because they live in private messages and personal clouds where no one is watching.
In our view, the warning about red-teaming is the most serious part of this story. If newer, more capable models from OpenAI and Google are getting worse at blocking nudes, then every upgrade risks widening the problem instead of shrinking it. Companies that ship these tools should treat blocking explicit outputs as a core feature, not an afterthought.
For readers, the practical takeaway is simple: do not assume that a famous AI company's guardrails will protect you or your family. The researchers say the content is already spreading unchecked, and regulators are not stepping in meaningfully. Until that changes, the burden falls on the companies building these models — and right now, the evidence suggests they are not doing enough.
