Grok ‘undressing’ backlash: why AI harms turn into platform governance fights

/ Technology / By

Summary: A backlash has erupted in the UK over the ability of Elon Musk’s Grok AI to generate image edits that effectively “undress” people. After criticism, X limited the feature so that only paying users can use it. UK ministers called the move “insulting” to victims of misogyny and sexual violence.

This isn’t a niche product controversy. It’s a preview of the next regulatory and platform governance fight: what happens when powerful generative tools make harassment cheap, scalable, and hard to trace.

What happened

From the BBC video explainer:

  • Grok AI was used to create edited images that digitally undress people.
  • Following backlash, X restricted Grok image editing so it’s available only to users who pay a monthly fee.
  • The UK government criticised the move as “insulting” to victims of misogyny and sexual violence.

Even without every technical detail, the shape of the problem is clear: a generative tool made it easy to create abusive sexualised imagery.

Why the paywall makes people angrier, not calmer

At first glance, “limit it to paying users” sounds like a control.

But it creates two bad signals:

  • Monetisation of harm: it looks like you’re charging for a capability widely viewed as abusive.
  • Misaligned incentives: if revenue comes from the feature, the platform has less incentive to eliminate it.

It’s similar to how some spam and fraud ecosystems work: a small group is willing to pay for capabilities that most users never want.

This is part of a larger category: non-consensual intimate imagery

Digitally “undressing” people sits in the same harm family as:

  • deepfake pornography
  • revenge porn
  • sexual harassment using synthetic media

The key element is non-consent.

The internet already struggles with this harm at human scale. Generative AI pushes it into industrial scale.

A model can be trained to follow rules (“don’t do X”), but:

  • it can be prompted around restrictions
  • it can generalise in unexpected ways
  • it can be fine-tuned or jailbroken

That means safety cannot rely only on “model behaviour.” It also requires:

  • product design constraints
  • detection and enforcement
  • user identity and traceability

The platform governance issue: where does responsibility sit?

When a tool enables abuse, responsibility often fragments:

  • “the user did it”
  • “the model just generates images”
  • “we restricted it behind a paywall”

Regulators are increasingly rejecting this buck-passing.

The likely direction of policy is:

  • platforms must show they designed systems to reduce foreseeable harms
  • not merely respond after outrage

What effective controls could look like

If a platform wants to demonstrate seriousness, the control stack typically includes:

  1. Hard capability limits
    Don’t allow certain transformations at all (e.g., nudification).

  2. Strong detection
    Detect and block generation of non-consensual sexualised imagery.

  3. Watermarking and provenance
    Make synthetic media easier to identify and trace.

  4. Reporting and rapid takedown
    Fast user reporting tools and dedicated enforcement.

  5. Meaningful consequences
    Account penalties that deter repeat abuse.

A paywall is not inherently a safety measure; it’s a distribution choice.

The cultural issue: “just a joke” isn’t a defence

A common pattern in online harms:

  • abusers frame it as humour
  • victims experience it as violation

Generative tools amplify this dynamic by reducing effort and increasing reach.

Why this is likely to escalate in 2026

Because:

  • generative tools are getting easier
  • image editing is becoming a default feature in platforms
  • victims’ images are widely available online

The combination makes abuse low-friction.

Bottom line

The Grok controversy is a warning that platform safety debates are moving from content moderation (what users post) to capability moderation (what tools can easily produce).

If platforms treat abusive synthetic imagery as a paid feature to be managed rather than a harm to be eliminated, governments will step in—and not gently.

Sources

n English