Defence’s AI Chief Reports an ‘Absolute Acceleration’ in Adoption as It Leans on ASD for Frontier Models
The Department of Defence has seen an “absolute acceleration” in its use of artificial intelligence over the past three months, its chief AI and information officer Chris Crozier told the Joint Select Committee on Artificial Intelligence on Friday, iTnews reported.
Crozier, Defence’s CIO since mid-2023, took on the chief AI officer role about 12 weeks ago through a direct appointment from the Defence secretary and the Chief of the Defence Force. He said the leadership was now aligned on AI’s impact and that, in his first three months, he and colleagues had interviewed more than 60 people across the department who were actively applying the technology, building a single portfolio of that work. His job, he said, is to make sure Defence understands the architecture and capabilities it is building for both warfighting and business productivity, and which results can scale.
On frontier models, Crozier said Defence currently relies on the Australian Signals Directorate to broker access, describing “exquisite capabilities” in those systems that are essential to national sovereignty and stressing the need to keep relationships with the companies that build them. He also said frontier models are not the be-all and end-all, arguing that a good open-weight model combined with an excellent harness, the software that connects a model to an organisation’s systems and data, can deliver a tremendous outcome, and that this is where Defence is focusing.
What it means in practice
Defence Innovation Review reported that the department’s submission to the committee said it may seek exemptions from whole-of-government AI rules where they clash with operations in contested, time-critical and classified settings, without naming which rules. Those rules require agencies to name accountable officials, keep registers of AI use, assess high-risk applications and publish transparency statements. Defence says human accountability remains central, especially in decisions about the use of force. The testimony adds to a running debate in Canberra over how government reaches the most capable systems, after earlier calls for embedded safety experts and early ASD access to new models and after Anthropic said it was open to reserving compute for Australian government priorities.
The committee is due to report by 30 November. What to watch is whether its recommendations address Defence’s request for carve-outs, how much of Defence’s AI portfolio is disclosed publicly, and whether the department leans towards open-weight models run on sovereign infrastructure as its AI work scales.