11. Live streaming offences pose unique legal and technical challenges for law enforcement and industry.
12. The speed and real-time nature of live streaming make it extremely difficult to proactively police interactions between the live streamer and the recipient. The practical effect of this is that it is harder for industry to deploy technology to detect, moderate or prevent live streamed child sexual abuse material. End-to-end encryption exacerbates this problem as it means the content of the communication cannot be accessed by industry or law enforcement.
13. On behalf of the NPCC, Chief Constable Bailey told us:
“the emergence of 4G and 5G and live streaming is going to present a greater risk … we know that there is a real problem in the area of the Philippines, and … I would have a real fear that with the emergence of 4G and 5G on the African continent, we are going to end up with a very similar situation”.[1]
14. We also heard evidence that, on occasions, law enforcement has difficulty in obtaining information about the online accounts of individuals suspected of grooming and live streaming. Commander Richard Smith, the professional lead for child safeguarding for the Metropolitan Police Service, told us about the live streaming of two girls aged six and nine who were being groomed to commit sexual acts. A number of offenders were watching and contributing to the grooming. The Metropolitan Police Service asked the service provider to remove the streaming and requested information which would identify the offenders. Commander Smith said that although the content was removed and the offenders’ accounts closed, the service provider:
“refused to provide any information regarding the offenders. While those offenders could no longer use their previous accounts to access the platform, there was nothing to stop them creating new accounts and to continue their previous offending. Without the police having access to data which might lead to the identification of offenders, [the Metropolitan Police Service are] unable to safeguard the children to whom offenders may have access.”[2]
15. When asked if Facebook knew the scale of live streaming on its platform, Ms Julie de Bailliencourt, Facebook’s Senior Manager for the Global Operations Team, explained that Facebook:
“don’t tend to look at prevalence of abuse across the content types but, rather, across the platform … whether it is a comment, a video or a photo, rather than specifically looking at live [streaming]”.[3]
She said that Facebook did not encounter “child safety specific streaming … on the platform too often”.[4]
16. Ms de Bailliencourt explained that concerns about the content of a live stream can be reported via Facebook’s reporting tools and that reports can be made as they are happening so that the reporter does not need to wait until the live broadcast is over. Facebook has a team of reviewers available 24 hours a day, seven days a week. She explained that, since late 2017, Facebook has been using machine learning to detect posts and live streams where someone might be expressing suicidal thoughts. When asked if such technology could be adapted to detect child sexual abuse, Ms de Bailliencourt said:
“this could offer really interesting opportunities on the child safety side. Although, again, as I have mentioned, because live streaming of child abuse is not a very common undertaking, thankfully, you know, this may provide limits to the learning that we may get from such reports.”[5]
17. Microsoft does not record figures about the number of specific live streaming offences reported to the National Center for Missing & Exploited Children (NCMEC)[6] but said that live streaming most commonly took place on Skype. Mr Hugh Milward, Senior Director for Corporate, Legal and External Affairs for Microsoft UK, said that this, in part, was the motivation behind Microsoft’s decision to fund the IWF research. He explained that based on the research:
“we quickly realised that, if you have one single live stream of abuse, that live stream is then captured and then shared on multiple times. And while it was – it is incredibly … difficult to stop that one instance of the live stream, that there must be a way … of developing technology that tries to address the way in which that live stream is then shared on multiple times.”[7]
It was this finding that “prompted us to focus more attention on to the development of PhotoDNA for video”.[8]
18. The collaboration between the IWF and Microsoft resulted in the development of PhotoDNA for Video. It is an example of the positive results that such cooperation can bring.
19. Google told us that, of all its products and services, YouTube was the platform most commonly used for the live streaming of child sexual abuse.[9] Users of YouTube can watch videos and upload their own videos to the platform. They can create a live stream via a webcam and other users can post comments or live chat as they watch the live stream. Google deploys its comments classifier to detect potentially inappropriate comments.[10] Those comments are then captured and removed and, if necessary, reported to NCMEC.
20. In relation to detecting child sexual abuse within the live stream itself, Ms Kristie Canegallo, Vice President and Global Lead for Trust and Safety at Google, told us that Google has “invested in technology that would allow us to monitor live streams and flag any potential inappropriate behaviour as well as flag whether minors are engaging in a live stream”.[11] In such cases, the live stream would be queued in a list pending review by a moderator.
“We have a dedicated team of human reviewers, that reply within minutes, to look at any live streams that are flagged and, to the extent that we saw CSAM there, we would terminate … that live stream … and then report it to NCMEC.”[12]
21. The live streaming of child sexual abuse is one of the most harmful forms of abuse that is affecting children today. Although it may be difficult to detect, the internet companies must demonstrate that they understand fully the scale of this abuse and are deploying sufficient resources to detecting this type of online-facilitated harm.