Incident·WS-0069·Dec 2025·in the wild·Child safety

Grok mass-produces sexualized images of real women and minors

xAI / Grok image editing on X

63/100
WS-4 Severe
Harm score 70, signal score 54. Autonomy: acted (the AI took a bounded action). How the scores work.
Harm 3/5
damage actually done
Reach 4/5
who was affected
Autonomy 2/4
how much the AI did on its own
Novelty 3/5
how new the behavior was
Permanence 4/5
how hard it is to undo

From late December 2025, X users widely prompted Grok to edit photos of real people into revealing or sexualized images, including apparent minors, often without the subjects' consent. CCDH sampled Grok's image posts from 29 Dec 2025 to 8 Jan 2026. It estimated about 3 million photorealistic sexualized images in those 11 days, roughly 23,000 of which appeared to depict children. On 9 Jan xAI limited image generation and editing to paying users. It triggered national blocks in Indonesia, Malaysia and the Philippines, an Ofcom investigation, a California AG cease-and-desist and an EU Digital Services Act probe.

Damage

~3.0M sexualized images in 11 days (95% CI 2.96–3.05M); ~23,338 apparently depicting children (95% CI 17,099–30,039), per CCDH extrapolation

Did they fix it?

reopened xAI's restrictions cut the flood but did not stop it - testers through October still got non-consensual sexualized images, and the UK, EU and Canadian cases are unresolved.

  1. Reported Jan 8, 2026

    From late December 2025, X users widely prompted Grok to edit photos of real people into sexualized images, including apparent minors. CCDH, sampling Dec 29-Jan 8, estimated about 3 million such images in 11 days, roughly 23,000 appearing… source

  2. Response Jan 14, 2026

    xAI restricted Grok image generation to paying users (Jan 9) and announced (Jan 14) that X users 'will no longer be able to use Grok to alter images of real people' into revealing clothing; it later restricted the Imagine feature to paid… source

  3. Claimed fix Oct 14, 2026 partial

    xAI called the problem an 'isolated' lapse and said it was urgently fixing 'lapses in safeguards'; it reported removing tens of thousands of accounts and filing tens of thousands of NCMEC reports in 2026, and (via talks with Ofcom) said… source

  4. Independent check Oct 14, 2026

    Independent testing repeatedly found the fixes incomplete. The Washington Post (Jan 15) found the standalone Grok app still undressed women; Euronews/AI Forensics (Jan 21) found explicit images still produced; Reuters (early Feb) ran… source

  5. Where it stands Oct 14, 2026 reopened

    After each announced curb, independent testers found Grok still producing non-consensual sexualized images; Ofcom's investigation has moved to assessing whether X breached its duties (no provisional finding or fine yet), and the EU DSA… source

The full case, with what would change the verdict →

Sources

CCDH, second source. CCDH figures are extrapolated from a 20,000-post sample scored by a classifier. CCDH could not tell whether consent was given or whether source images were already sexualized. The California AG cited a separate analysis: more than half of 20,000 images generated between Christmas and New Year showed people in minimal clothing. Consequences are listed separately: country blocks, Ofcom, California AG and EU DSA. Baltimore also sued xAI on 24 Mar 2026 (Guardian, not independently fetched).

Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.