X is restricting Grok's image capabilities via the @grok command for free users, but they can still create sexual deepfakes via X's “Edit image” or Grok's app
X's sexual deepfake machine is still running, despite Grok saying otherwise.
The VergeRobert Hart
Context & Ripple Effects
Earlier coverage had documented people using Grok to alter images of real women into abusive and sexualized depictions, making this a question of whether safeguards work across every access route rather than at a single prompt surface. The reported free-tier limit leaves the image editor and standalone app as alternate paths.
The subsequent ban on editing real people into revealing clothing and later tests that found Grok still processed many requests to undress men underline the enforcement problem: policy changes must be applied consistently across product interfaces.
First-order effects
Free users lose image-generation access through the @grok command, while users of X's image-editing feature and Grok's app retain reported routes for creating sexual deepfakes.
X and Grok face an immediate trust and moderation gap because the restriction changes one entry point without removing the reported harmful capability from the broader product set.
Second-order effects
Safety enforcement shifts from a command-level control to a cross-surface product problem: X must align prompts, image editing, and the Grok app or users can simply migrate to the least-restricted interface.
The gap makes later safeguards easier to test and challenge; the reported continued ability to undress men after an update shows why narrowly framed restrictions may not resolve broader misuse concerns.
Third-order effects
If AI image tools remain available through inconsistent interfaces, effective governance will increasingly be judged by end-to-end abuse resistance rather than by stated policy or isolated feature limits.
The case points toward tighter expectations around consent and access controls for image-editing systems, especially where platforms combine social distribution with generative tools; the eventual regulatory response remains uncertain.
The trend: Generative-image platforms are moving from broad feature rollout toward the harder task of enforcing consent-based safeguards consistently across every user surface.
You may have seen reports swimming around claiming that X finally restricted the Grok deepfake/undressing features behind a paid subscription. — That simply isn't true. Not even close. www.theverge.com/news/859309/ ...
What's happened here is Musk has made the choice to turn off safety systems. — Normally, these models contain instructions not to generate harmful outputs, Musk has simply disabled these. — This a choice and, once again, Musk has not been held accountable. Is he ever held to…
Senators Ron Wyden, Ben Ray Lujan, and Ed Markey just published an open letter urging Apple + Google to remove the X and Grok apps from their app stores: — www.nbcnews.com/tech/interne... [embedded post]
After further testing, Grok will still produce nonconsensual sexual images in the Grok tab on X, on the Grok website, and in the Grok app, even for non-paying users. The only pullback has been on the capabilities of the @grok reply bot. https://www.nbcnews.com/...