This month, I’ve been participating in hearings of the Joint Select Committee on Artificial Intelligence, questioning AI developers, business groups, civil society organisations and academics about how Australia should navigate the rapid advent of AI. I have focused many of my questions on AI safety.
AI promises many real benefits, from advancing medical research to increasing productivity, but these benefits depend on trust – they will only be realised if we get the safeguards right so Australians can take up this new general purpose technology with confidence.
We’ve heard a lot in recent weeks about potential existential risks – misuse or misalignment – and it can seem like all the relevant decisions will be made overseas. But there are steps Australia can take now. So far, the hearings have confirmed many of the policy asks included in my May 2026 policy paper, Shaping Our Future of AI for Australia.
The simplest step we can take now is to significantly increase investment in Australia’s AI Safety Institute (AISI). This is our first line of defence in a rapidly changing world and its mandate is broad - testing new models, working with international counterparts to set global standards and advising Government regulators on preparing for the novel risks AI presents. So far not a single witness has said that the $8 million in annual funding for the AISI is adequate. A $100 million investment would give us a credible global voice.
We also need to know quickly when things go wrong. Last month the Prime Minister revealed that, during an internal test, an OpenAI agent got around security blocks on a Services Australia website to access Medicare statistics. The hack happened on 18 June, but the Australian Government wasn’t told until 10 September, via an email sent to a public inbox. This is a woefully inappropriate way to report such a major incident.
We only know about this because OpenAI chose to fess up. Right now, AI companies are not legally required to report incidents like this to the Government. My May 2026 proposal for a mandatory transparency and incident reporting scheme for AI developers has been backed by many witnesses, including OpenAI’s Chief Strategy Officer Jason Kwon.
Biosecurity risks are about to increase, with emerging models dramatically increasing the number of people who could feasibly build a biological weapon. We could build on our strong global biosecurity reputation by requiring screening of any imported synthetic DNA material. This has strong expert support and would not require new legislation.
When I surveyed Curtin residents on AI, eight in ten said the Government isn’t doing enough. Australians support getting ahead of the risks now rather than scrambling later. This month’s hearings showed that experts agree.
I’ll keep pushing the Government to act rapidly on these and other safeguards, so we can confidently realise the benefits AI offers.