Submission 113
Sovereign Australia AI
Submitter group: Companies.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Argues Australia needs greater sovereignty across the AI stack, particularly in models, to avoid dependence on foreign providers, citing economic, geopolitical, defence, talent, cultural, democratic, resilience and legal reasons and a six-layer definition. The author is CEO of Sovereign Australia AI, which says it is building an Australian-built language model.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that most corporate and government AI use relies on foreign models, sending spending offshore with little benefit to Australia.
The vast majority of corporate and government use of AI leverages foreign AI models. Every token consumed funnels money overseas (predominantly to the USA) with little to no benefit to Australia.
Argues that the AI supply chain is controlled by US executive discretion, citing episodes it says have already affected Australian access.
Both episodes show the AI supply chain is controlled by US executive discretion, and that discretion has already been exercised against Australia.
Argues that defence teams use foreign-built models without knowing their training data, biases or hidden vulnerabilities.
However, our military are expected to protect our nation using models and tools built by foreign non-military companies. Our defence teams do not know the source material used to train the model, any biases introduced in training or any hidden trojans that may lurk in the models and tools.
Argues that Australian data makes up about 1 to 1.5 per cent of frontier models, so Australian-centric answers need extra prompting and are more likely to be inaccurate.
Australian data makes up approximately 1%-1.5% of existing frontier AI models. This means that while we can extract an Australian-centric answer from one of these models, it takes additional prompting steps, and the answer has a higher chance of being inaccurate due to confabulation (hallucination).
Argues that the three layers of AI sovereignty beyond infrastructure and applications are the most critical and lack sovereignty at scale.
While there is solid focus on building sovereign compute (the infrastructure layer) and developing sovereign user applications, the three remaining layers (considered the most critical layers by this author) lack sovereignty at scale.
Argues that critical personal data should be stored domestically through a provider not subject to the US CLOUD Act.
Further, critical personal data should only be stored domestically through a provider not subject to the US CLOUD Act.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Invest in retaining AI talent by making Australia an attractive place to build sovereign capability.
We must invest in talent retention by making Australia an attractive place to build sovereign capability.
Build Australian models with far greater Australian content, drawing on digital public records and state library datasets.
We need Australian built models with far greater Australian content from data troves that exist today such as digital public records, unique data sets held by state libraries and more.
Develop Indigenous language capability in Australian AI models.
Australia needs to have indigenous language capability (as difficult as that task may be) in our AI models.
Mandate that AI used by key industries and government complies with Australian law and keeps Australian data onshore.
AI used by key industries and all of government should have a mandate to utilise AI that can clearly comply with all laws and ensure that Australian data never goes offshore for storage or processing, even at rest.
Require that critical personal data be stored only domestically through a provider not subject to the US CLOUD Act.
Further, critical personal data should only be stored domestically through a provider not subject to the US CLOUD Act.