Tech companies argue that upcoming Australian safety standards may inadvertently increase challenges for generative AI systems in identifying and preventing online child abuse and pro-terrorism material. The proposed standards by the eSafety commissioner, Julie Inman Grant, require providers to detect and remove such content “where technically feasible” and deter new material. The regulations encompass various technologies, including websites, cloud services, text messages, and generative AI models.
WeProtect Global Alliance, a non-profit consortium combating online child exploitation, supports the standards, emphasizing the risk of open-source AI generating illegal content. However, major tech firms like Microsoft, Meta, and Stability AI express concerns. They contend that the standards, as drafted, might impede the effectiveness of AI safety models designed to detect and flag inappropriate content. Microsoft warns against overly strict criteria for training data, while Stability AI cautions that broad definitions could complicate compliance.
Meta, Facebook’s parent company, highlights the difficulty of enforcing safeguards once an AI tool is downloaded. Google suggests excluding AI from the standards and incorporating it into the broader government review of the Online Safety Act. The companies, including Apple from a previous week, stress the need for standards to ensure scanning cloud and message services “where technically feasible” doesn’t compromise encryption.
In response, Commissioner Inman Grant clarifies that the standards won’t mandate breaking encryption or indiscriminate scanning of personal data. She acknowledges ongoing discussions about potential amendments and emphasizes that industry should share responsibility for tackling illegal content on their platforms. The final standards are expected to be presented in parliament later this year.
Commissioner Inman Grant further states that the proposed standards are not intended to compel the industry to weaken encryption, monitor texts, or conduct indiscriminate scans of extensive personal data. She emphasizes that the eSafety commissioner is actively considering potential amendments to ensure clarity on this matter.
In defense of the standards, Inman Grant asserts that eSafety does not believe the industry should be exempt from its responsibility to combat the hosting and sharing of illegal content on their platforms. She notes that some major end-to-end encrypted messaging services are already taking measures to detect harmful content.
Despite the concerns raised by tech companies, the eSafety commissioner’s commitment to addressing potential issues and considering amendments reflects an ongoing effort to strike a balance between ensuring online safety, especially for children, and not unduly burdening the development and deployment of advanced AI technologies.
As the final versions of the standards are slated to be presented in parliament for consideration later this year, the ongoing dialogue between regulatory bodies and tech companies will likely play a crucial role in refining the standards to meet both the objectives of child safety and the legitimate concerns of the technology industry. Balancing the need for robust online safety measures with the responsible development and deployment of AI technologies remains a complex and evolving challenge for regulators and tech companies alike.
