‘Damaging content’: Big tech short on answers and willpower in fight against evil child sexual abuse material

Google and Apple are among tech giants putting profits before child protection when it came to combating the explosion of deepfakes and other sexual abuse material, Australia’s eSafety Julie Inman Grant claims. She said companies weren’t answering key questions she’d put to them and weren’t employing the full arsenal of tools to fight the depraved crimes occurring on their services, even though they had gone to the effort of creating them. “They’ve developed some very amazing technologies,” she said on Saturday. “Apple with NeuralHash and Google with Content API, they’re using on some of their services but by no means all of their services.” Ms Inman Grant said AI technologies allowed these companies to “scale more rapidly and with more accuracy, but they’re choosing not to use them”. “They are not using then on YouTube. They are not using them on iCloud, (and on) elements of iMessage, they’re not using on some of the mailing applications and none of them are using them on live-streaming,” she said. She believed their reasons boiled down to costs. “There is added expense,” she said. “It may slow down the live streaming to a certain degree but again I would expect that these technology companies would be able to scale that. “But what’s more concerning to me is (big tech companies) are not willing to (answer) really basic safety questions, fundamental questions like ‘how many trust and safety personnel do you have?’ “If you’ve got job descriptions and a payroll system you can tell me how many people are working on trust and safety. And the simple fact of the matter is that once Elon Musk came and took over Twitter and turned it into X . . . they cut up to 80 per cent of their trust and safety and public policy people which isn’t a very good signal to shareholders or for brand safety and advertisers.” She said other tech giants followed X’s lead. “Once the band-aid was ripped off there was a much more quiet firing. Meta (owner of Facebook, Instagram, WhatsApp and Messenger) started first,” she told an ABC podcast. “I wrote them a letter and basically said we want to make sure you’re keeping your trust and safety (staff) and your local Australian representatives here because we need to make sure that we are able to communicate with them about regulatory concerns, so that slowed it a little.” In one example, she cited that in June, The New York Times revealed YouTube, the world’s largest video platform, had told its content moderators to favor “freedom of expression” over the risk of harm in deciding what to take down. “When a company is trying to argue to the Australian Government for instance that it is educational and safer than other platforms it’s hard ... when you’re rolling back those trust and safety personnel and the policies to keep more harmful content there,” she said. Ms Inman Grant said she believed Australia had reached crisis point with deepfake technology, which had advanced at an incredible pace. “In 2020, we started our tech trends and challenges program and we did a brief on deep fakes and at that time, just five years ago, it took a lot of computing power, a high degree of technological expertise and a lot of images to create a credible deep fake,” she said. “That’s why we started seeing celebrities being targeted initially. “But today the technology is so accessible and virtually free that a teenage boy can download a free nudifying or declothing app and just harvest one or two pictures of a classmate and turn it into a credible deepfake which is incredibly harmful and devastating for the target.” “We have sent letters to every Education Minister and Secretary and we have developed a deepfake image-based abuse incident management plan so that schools know how to deal with this, when to call the police, when to report to eSafety. “We have an image-based abuse scheme where we have a 98 per cent success rate in terms of getting this content taken down and one of our areas of enforcement focus is around declothing and nudifying apps or other AI apps that are being used to generate synthetic child sexual abuse material.” With AI generally, Ms Inman Grant said, as with any technology, the challenge was how to harness the benefitsw, while minimising and mitigating the risks. “Initially when we saw that AI drag-race happening in 2023, which very much looked like the social media era of moving first and breaking things, let’s just get to market first, get market share, it really concerned me. “Now that we’re having debates around AI sovereignty and developing AI quickly I think there’s a fallacy that if you look at applying appropriate guardrails for really high risk situations that you are going to undermine innovation and development. “I would like to see us harnessing AI as a way of encouraging safety by design. The truth is if companies like Anthropic are mindfully choosing the data that they’re scraping, how they’re training it, who is training it for them, how they’re deploying it, how they’re red teaming and testing it, we could have really significant productivity gains but minimise the risks. “And the risks right now are fairly significant when it comes to image-based abuse and the sexual privacy that we’re looking at, but also in terms of AI sexualised chatbots and companions.” “We are in the process of assessing industry developed codes that will potentially cover AI chatbots,” she said. “(If companies don’t provide) adequate guardrails for AI companions and chatbots, preventing 12 year old’s from accessing really potentially damaging content and forming relationships with chatbots that will direct them to engage in specific sexual acts, then I will not register these and I will make standards.” “We now have some new powers around codes and standards. With the passage in November of the Social Media Minimum Age Bill means that codes and standards violations I can fine these companies up to $49.5 million for violations. We have just put together an enforcement taskforce and are looking at the platforms that we are going to target, but we will end up playing a game of whack-a-mole. “We are just one small country with a relatively small regulator, if you look at OFCOM which is the UK sister regulator of ours, they just set up their Online Safety Act in 2023 and already have double the staff. They have 500 people to our 250. “To take on not only this many powerful companies but if you think of the broader universe of websites and dating sites and gaming platforms and wily perpetrators, we really need a lot more global regulatory coherence and partnership to tackle this.” She wasn’t sure a specific new legislation like Europe’s AI Act was needed just yet. “I don’t think we are there yet,” she said. “I just met with my fellow digital regulators from the ACCC, the ACMA and the Privacy Commissioner last week. We are trying to harness the regulatory tools that we already have, but regulating technology and regulating the internet is extremely tricky business particularly because most of the purveyors and profiteers are based overseas. “We are also seeing a very different approach to AI regulation in the United States, which is mostly hands-off. “But what I would say is that the Take it Down Act which was just signed into law pretty much replicates our image-based abuse scheme and what we’re doing with deepfakes, so I would like to be able to be working with the US on similar proposals where we can find common ground, because when you think about online safety policy or regulation it’s about where you draw the line in terms of preventing harm.”

‘Damaging content’: Big tech short on answers and willpower in fight against evil child sexual abuse material

The No. 1 source of news for every West Australian

Everyday Digital

Everyday Digital + Weekend Papers