My Lords, I will speak to all the amendments in this group, other than Amendment 295 from the noble Baroness, Lady Jones. Without stealing her thunder, I very much support it, especially in an election year and in the light of the deepfakes we have already seen in the political arena—those of Sadiq Khan, those used in the Slovakian election and the audio deepfakes of the President of the US and Sir Keir Starmer. This is a real issue and I am delighted that she has put down this amendment, which I have signed.
In another part of the forest, the recent spread of deepfake photos purporting to show Taylor Swift engaged in explicit acts has brought new attention to the use, which has been growing in recent years, of deepfake images, video and audio to harass women and commit fraud. Women constitute 99% of the victims and the most visited deepfake site had 111 million users in October 2023. More recently, children have been found using “declothing” apps, which I think the noble Baroness mentioned, to create explicit deepfakes of other children.
Deepfakes also present a growing threat to elections and democracy, as I have mentioned, and the problems are increasingly rampant. Deepfake fraud rates rose by 3,000% globally in 2023, and it is hardly surprising that, in recent polling, 86% of the UK population supported a ban on deepfakes. I believe that the public are demanding an urgent solution to this problem. The only effective way to stop deepfakes, which is analogous to what the noble Baroness, Lady Kidron, has been so passionately advocating, is for the Government to ban them at every stage, from production to distribution. Legal liability must hold to account those who produce deepfake technology, create and enable deepfake content, and facilitate its spread.
Existing legislation seeks to limit the spread of images on social media, but this is not enough. The recent images of Taylor Swift were removed from X and Telegram, but not before one picture had been
viewed more than 47 million times. Digital watermarks are not a solution, as shown by a paper by world-leading Al researchers released in 2023, which concluded that
“strong and robust watermarking is impossible to achieve”.
Without measures across the supply chain to prevent the creation of deepfakes, the law will forever be playing catch-up.
The Government now intend to ban the creation of sexual imagery deepfakes; I welcome this and have their announcement in my hand:
“Government cracks down on ‘deepfakes’ creation”.
This will send a clear message that the creation of these intimate images is not acceptable. However, this appears to cover only sexual image deepfakes. These are the most prevalent form of deepfakes, but other forms of deepfakes are also causing noticeable and rapidly growing harms, most obviously political deepfakes—as the noble Baroness, Lady Jones, will illustrate—and deepfakes used for fraud. This also appears to cover only the endpoint of the creation of deepfakes, not the supply chain leading up to that point. There are whole apps and companies dedicated to the creation of deepfakes, and they should not exist. There are industries which provide legitimate services—generative Al and cloud computing—which fail to take adequate measures and end up enabling creation of deepfakes. They should take measures or face legal accountability.
The Government’s new measures are intended to be introduced through an amendment to the Criminal Justice Bill, which is, I believe, currently between Committee and Report in the House of Commons. As I understand it, however, there is no date scheduled yet for Report, as the Bill seems to be caught in a battle over amendments.
The law will, however, be extremely difficult to enforce. Perpetrators are able to hide behind anonymity and are often difficult to identify, even when victims or authorities are aware that deepfakes have been created. The only reliable and effective countermeasure is to hold the whole supply chain responsible for deepfake creation and proliferation. All parties involved in the AI supply chain, from AI model developers and providers to cloud compute providers, must demonstrate that they have taken steps to preclude the creation of deepfakes. This approach is similar to how society combats—or, rather, analogous to the way that I hope the Minister will concede to the noble Baroness, Lady Kidron, society will combat—child abuse material and malware.