Social Media: Non-consensual Sexual Deepfakes

Baroness Keeley Excerpts

Wednesday 14th January 2026 Lords Chamber

[Baroness Lloyd of Effra (Lab)]

These are serious offences and the noble Lord is right that there is consensus on this. The decisions on prosecution and sentencing are for the police and courts. They should know that, as we have said, they have our support in taking that action.

[Baroness Keeley (Lab)]

My Lords, I welcome the Government’s announcement that they are bringing legislation into force this week to tackle this issue, and I welcome the news that Ofcom has launched a formal investigation to determine whether X has complied with its duties under the Online Safety Act. Ofcom should act urgently on this. I also support the Government’s intention to act on gaps identified in our online safety legislation, such as the fact that not all chatbots are covered.

I commend my noble friend Lady Owen of Alderley Edge for her work on banning deepfakes through amendments to the Data (Use and Access) Bill, and her excellent continuing work on this issue. There is an issue about the lack of transparency in how chatbots such as Grok are trained. As I understand it, if an image or multimodal model can generate non-consensual sexual imagery or deepfake pornography, it is certain that the model was trained on large, uncurated web scrapes where such material is common. Does my noble friend the Minister agree that this gives new impetus to the Government tackling the issue of transparency in the training of AI models, which is a matter we are looking at on the Communications and Digital Select Committee in terms of the transparency needed to deal with issues of AI copyright? This is a new and very pressing part of that issue.

[Baroness Lloyd of Effra (Lab)]

My noble friend raises good questions about training and testing. As she will also know, we are bringing forward measures in the Crime and Policing Bill that will allow testing in certain narrow circumstances, so that developers can make sure that the models they bring forward are not able to disseminate these kinds of awful images or CSAM. These are very important things and we are working very carefully with others to find the right regime for these models.