Hugging Face used to strip women and children - the cost of open weights is on the platform now

A report documents how Hugging Face is used to easily create non-consensual intimate images - including of children. The cost of open weights is no longer theoretical.

In plain terms - A report from The Verge documents how Hugging Face is being used to easily create non-consensual nude imagery - including of children - and that the platform is doing very little to prevent it. The cost of open weights is not theoretical; it is on the platform, now.

The fact

According to The Verge coverage on July 28, 2026, the Hugging Face platform - a major hub for open-source models - hosts and redistributes models used to easily generate "undressed" images of women and children without their consent. The Verge explicitly writes that Hugging Face "isn't doing much to prevent" the output of sexualized deepfakes generated by the hosted models. The article fits into the broader debate opened by the OpenAI/Hugging Face breach in July (post #1460, frontier-access-control thread), but on another side: the adverse use of public artifacts, not the model incident.

Our reading

Three tensions to hold at the same time. One: open weights do not have built-in "content moderation" - moderation, when it exists, is played at the hub level (removal of models), not at the model level itself. Two: the documented class of use (non-consensual intimate imagery, CSAM) is already illegal in most jurisdictions - this is not primarily an AI policy problem, it is a problem of effective enforcement at scale. Three: Hugging Face is not designed as a content moderator in the legal sense, but the scale makes inaction untenable - it is the paradox of the neutral hub.

To watch

• HF response: explicit removal policy, third-party audit of risky models.
• Regulator action (Ofcom UK, DSA in Europe) likely to target HF directly.
• Technical countermeasures (mandatory watermarking of outputs, model provenance).