Submission 259
Gradient Institute
Submitter group: Research institutes.
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.
Says it is optimistic about AI's benefits, which it argues will only be realised if AI is developed and used safely. Recommends substantially increased funding for the Australian AI Safety Institute, mandatory AI incident reporting with the Institute as receiving body, and a dedicated AI venture fund for safe and responsible AI technologies.
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 the benefits of AI will only be realised if it is developed and used safely and responsibly.
However those benefits will ONLY be realised if AI is developed and used safely and responsibly.
States that the Australian AI Safety Institute's budget is about $7.5 million a year, roughly one-sixteenth of the UK institute's.
Yet Australia’s AISI has a budget of $29.9 million over four years so around $7.5 million a year. The UK's AI Security Institute operates on roughly $120 million a year - approximately 16 times the funding of the Australian AISI.
Argues that a mandatory incident reporting regime should cover serious harms and significant near-misses with clear thresholds and timelines.
mandatory incident reporting for advanced AI systems, with the AI Safety Institute as the receiving body. We agree. A regime should cover serious harms and significant near-misses, set clear thresholds and timelines, and build a national picture of AI harms that regulators can act on.
Argues interactions between organisations' agents create cascading errors, unintended collusion and unsanctioned coordination that no single organisation can fully observe.
when agents from different organisations interact, new failures emerge from the interactions themselves: errors that cascade between agents, agents converging on collusive behaviour nobody instructed, and coordination through channels no one sanctioned, as occurred in the OpenAI/Hugging Face incident. No single organisation can see or control the whole system, so no single organisation will observe the pattern.
Argues that Australia should build critical AI system components beyond the largest frontier models, serving demand for safe, secure and trustworthy AI.
Australia is very unlikely to train the world's largest frontier AI models but there are many other critical components of AI systems that Australia can and should be building. The global AI market is looking for technologies and products that make powerful AI safe, secure and trustworthy enough to deploy at scale.
Recommends the Committee consider a dedicated AI venture fund to co-invest alongside private capital in safe and responsible AI companies.
We recommend the Committee consider a dedicated Australian AI venture fund to co-invest alongside private capital in companies that build safe and responsible AI technologies.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Raise Australian AI Safety Institute funding to a level comparable to the UK's relative to economy size, with a clear mandate and technical resources to research and test frontier and agentic AI.
We recommend that the Australian AISI's funding be raised to a level comparable, relative to economy size, with the UK AISI, with a clear mandate and the technical resources to conduct research and test frontier and agentic AI systems.
Introduce mandatory incident reporting for advanced AI systems to the AI Safety Institute, covering serious harms and significant near-misses, with clear thresholds and timelines and a national picture regulators can act on.
Global Shield Australia has argued for the same principle in AI: mandatory incident reporting for advanced AI systems, with the AI Safety Institute as the receiving body. We agree. A regime should cover serious harms and significant near-misses, set clear thresholds and timelines, and build a national picture of AI harms that regulators can act on.
Consider an Australian AI venture fund co-investing with private capital in safe, responsible AI companies, helping them grow locally while targeting global markets.
We recommend the Committee consider a dedicated Australian AI venture fund to co-invest alongside private capital in companies that build safe and responsible AI technologies. Such a fund would help such companies grow here rather than relocate while still being focussed on a global market.
Focus the dedicated AI venture fund on building, retaining and commercialising Australian capability in AI safety, security, assurance and governance software infrastructure.
Back an Australian industry building safe and responsible AI technologies. Establish a dedicated AI venture fund to build, retain and commercialise Australian capability, with a focus on areas where safety and trust are the outcome: AI safety, AI security, assurance, and the software infrastructure that governs AI systems