Grok makes sexual deepfakes, three countries block it

The images showed real people who could be recognised, mostly women and also minors. The harm was immediate and hit individuals. Victims are "left with no way to prevent the spread" of the images, said Baltimore mayor Brandon Scott.

Indonesia blocked Grok on 10 January, as the first country to do so. Malaysia followed a day later, the Philippines on 16 January. The blocks were temporary. By the end of January the three countries had restored access, after X and xAI put extra safeguards in place.

Sexual images of identifiable people, made without their consent. With this kind of harm the victim can be named, the damage is personal and hurts their reputation, and no update to the model can undo it afterwards.

I see this as a different kind of problem from the Hugging Face intrusion. With Grok, the system did what users asked, and the harm came from the request. At Hugging Face, the agents did something nobody had asked for.

Public debate about AI usually puts both under one word, risk. They need different solutions, and the people opposing those solutions are different too.