Contribution
My Lords, I rise today in support of Amendment 203 in the name of the noble Baroness, Lady Kidron. I declare an interest as a recent guest of Google at its Future Forum policy conference. I apologise for not being able to make Second Reading and for not being present for my last amendment; as a newer Peer, I am very new to this and still learning as I go. I am very grateful to the noble Baroness, Lady Kidron, for stepping in.
I commend the wording of the noble Baroness’s amendment, which tackles the full process of training these models, from the collection of data or images to use as training data, all the way through to possessing a model. With these apps easily downloadable on app stores, there is a lack of friction in the process. This means that we have seen horrific cases of children using these apps in schools across the world with devastating consequences. In summer, I met the father of a little girl who had been bullied in this way and sadly took her own life.
I am very grateful to the noble Baroness, Lady Kidron, for this thoughtful and comprehensive amendment, which seeks to future-proof with its inclusion of avatars. We have already seen these threats evolving in the metaverse. I encourage the Government to adopt this amendment so that we can begin to see an end to this abusive market.
I turn to my Amendment 211G. I am very grateful to the noble Lords, Lord Clement-Jones and Lord Browne of Ladyton, and the noble Baroness, Lady Kidron, for putting their names to it. Noble Lords may recognise it from my Private Member’s Bill on non-consensual sexually explicit images and videos. I will keep my remarks brief as many of your Lordships were present on Friday.
The amendment seeks to create offences for the non-consensual creation of sexually explicit content and to close the gaps in the Sexual Offences Act. It is, vitally, consent-based, meaning that victims do not have to suffer the trauma of proving the motivation of their perpetrator. It includes solicitation to prevent any creation laws being circumnavigated by asking those in other jurisdictions to create such content for you through the uploading of clothed images to forums. Finally, it includes forced deletion so that victims can clearly see their rights to have the content destroyed from any devices or cloud-based programmes and do not have to live in fear that their perpetrator is still in possession of their content.
This amendment is inspired by the lived experience of victim survivors. The Government have repeatedly said that they are looking for the most suitable legislative vehicle to fulfil their commitment to criminalise the creation of sexually explicit deepfakes. It seems they did not think my Private Member’s Bill was the right vehicle, but it is my firm belief that the most appropriate legislative vehicle is the one that gets there quickest. I am hopeful that the Government will be more receptive to an amendment to their legislation, given the need urgently to tackle this rapidly proliferating form of abuse.
Amendment 211H addresses the problem of sexually explicit audio, which the noble Baroness, Lady Gohir, spoke about so movingly in Friday’s debate. We have seen satirical voice cloning, such as of Gareth Southgate at the 2024 Euros. However, the most state-of-the-art systems now require around three seconds of voice audio data to create speech on a parity with a human. This could be data from a short phone call or a TikTok video. As we are reaching the point where less data is required to create high-quality audio, this now has the potential to be weaponised. There is a real risk that, if we do not future-proof against this while we have the opportunity, it could rapidly develop in the way that sexually explicit deepfake images have. We are already seeing signs of new sexually explicit audio online. Its ease of use combined with its accessibility could create a huge risk in future.
Henry Ajder, the researcher who pioneered the study of non-consensual deepfake image abuse, said:
“2024 has seen AI generated voice audio widely used in spreading political disinformation and new forms of fraud, but much less attention has been paid to its potential as a tool for digital sexual abuse”.
In his research in 2018, he observed several cases of online communities experimenting with voice-cloning capabilities, targeting celebrities to create non-consensual “synthetic phone sex” content. This Bill could be a key opportunity to future-proof against this problem before it becomes widespread.