Assistant Science and Technology Minister Andrew Charlton warned this week that powerful AI models "are already doing things their creators never intended: cheating, deceiving, going their own way." His remarks came as the Australian government launched an AI Safety Institute with a four-year budget of A$29.4 million, a move that signals Canberra is taking AI risks seriously but may fall short of what the moment demands.
Harms already hitting Australians
Australians are already experiencing direct harm from AI. Nudify apps, sophisticated scams, deepfakes, voice cloning, and chatbots that have isolated teenagers or encouraged self-harm are in circulation. Security agencies report rising cybersecurity threats tied to AI capabilities.
Charlton told the AI Safety Forum at the University of Sydney that advanced models are doing things their creators never intended. He said rapidly expanding capabilities and agentic AI systems lack reliable controls. The United Nations' Independent International Scientific Panel on AI recently raised concerns about global concentration of power, resources, AI capability, inequality, and the technology's impact on how people think, reason and work.
Inside the AI Safety Institute
Kate Conroy, a philosopher and Royal Australian Air Force reservist who leads the institute, outlined its three-part remit. The institute analyzes and tests new AI models, supports government regulators and agencies in responding to emerging AI capabilities, risks, harms and trends, and shapes safe AI development, deployment and international governance in Australia's interests. The institute is already collaborating with the Gradient Institute on multi-agent risks - scenarios where AI agents interact without supervision, producing unpredictable outcomes - and with CSIRO on AI alignment, the science of ensuring AI systems act according to users' values and goals. Conroy said the institute will tackle both immediate harms affecting Australians, with particular focus on the most marginalized and vulnerable, as well as frontier risks and harms.
A budget that lags global peers
The institute's A$29.4 million over four years is a fraction of what comparable nations spend. The United Kingdom allocated about A$460 million in its 2025 spending review to its AI Security Institute. Singapore's AI Safety Institute has an annual budget of around A$11 million, and Canada set aside A$50 million over five years.
Industry spending makes government budgets look tiny. OpenAI alone spent US$19 billion on research and development in 2025. At the forum, proposals to significantly increase the Australian institute's budget drew strong support.
Beyond the institute: legislation and inclusion
Two practical next steps emerged at the forum: legislate a digital duty of care to help curb AI-supercharged online harms, and address digital exclusion to ensure all Australians can benefit from the technology. The Tech Policy Design Institute's recent assessment highlights Australia's strengths but also gaps, including securing computing power for research and public interest activities. Without that, the public sector and civil society risk falling behind.
Multiple speakers at the forum said Australia has a perhaps unique opportunity to make a difference. The country has skilled AI safety researchers and experts who study social impacts, and there is a moral imperative to address the risks and opportunities of AI for all Australians.
Why this matters for Government
The AI Safety Institute represents a structural response, but its effectiveness depends on legislative backing, sustained funding, and skilled public servants who can interpret AI capabilities and risks. Professionals in policy, regulation, and technology delivery can build relevant skills through programs like the AI Learning Path for Policy Makers, which covers AI governance and data-driven decision-making. For ongoing analysis of how AI is reshaping public sector work, the AI for Government resource hub offers case studies and policy guidance.
Your membership also unlocks:
