Meta has warned the Australian government that proposed privacy law changes could hinder its AI development, arguing that personal data from Facebook and Instagram posts is essential for training models to understand Australian concepts. In a submission to the Productivity Commission, the tech giant said generative AI requires large, diverse datasets that reflect how people discuss culture, art, and trends.
Meta stated that synthetic data or legislative texts are insufficient for training its AI, Llama, which has been learning from public posts since last year. The company emphasised that human discourse on its platforms provides vital learning about Australian realities and figures, and that effective AI products depend on such authentic training data.
While Meta offers European users an opt-out option due to specific legal frameworks, Australian users currently lack this choice. The company expressed concern that Australia's privacy reforms could become out of step with international norms, potentially discouraging AI investment and conflicting with other digital policy objectives.
Other companies, including Bunnings and Woolworths, also criticised the proposed changes. Bunnings, appealing a privacy ruling on facial recognition, argued that privacy must balance with workplace safety. Woolworths warned that reforms could hinder personalised customer experiences. Google echoed calls for clearer AI regulations and copyright law changes to allow training without infringement.
