X's Grok Still Generates Nonconsensual Deepfakes

After a wave of nonconsensual sexual deepfakes circulating on X, the company’s @Safety account posted changes limiting the Grok account’s ability to edit images of real people in revealing clothing. According to X, Grok will censor prompts like “put her in a bikini,” restrict image creation and editing via the Grok account to paid subscribers, and apply geoblocks in jurisdictions where generating such images is illegal.

However, The Verge tested the updated controls and found they don’t fully stop abuse: reporters were still able to generate revealing edits (including a bikini image) with a free account. X and xAI owner Elon Musk attributed issues to “user requests” and occasional “adversarial hacking of Grok prompts.” Regulators are responding — Ofcom has opened an investigation and the UK is bringing into force a law criminalizing nonconsensual intimate deepfakes — and Prime Minister Keir Starmer said X must ensure compliance. The Verge notes the company’s policy announcements have not, as of its testing, eliminated the problem.