Australia should require independent experts to scrutinise powerful artificial intelligence systems from inside the companies developing them, according to early recommendations from a bipartisan parliamentary inquiry.
The committee examining AI has recommended independent safety checks, greater access for Australian security agencies and stronger reporting requirements following a series of incidents involving autonomous AI systems.
Under the proposal, independent experts could be embedded within leading AI companies and given access to models, safety systems and internal testing, with an obligation to report serious problems to the government.
The Australian Signals Directorate and other security agencies could also receive early access to frontier AI models as a condition of companies training advanced systems in Australia.
Independent oversight proposed
The committee has also recommended reserving part of Australia’s rapidly expanding computing capacity for national security, universities, healthcare and other areas considered strategically important.
The push reflects concern that Australia risks becoming primarily a consumer of AI developed overseas while supplying the energy, land and infrastructure needed to operate it.
Australia is attracting billions of dollars of data centre investment, but much of the underlying technology remains controlled by US-based companies.
The inquiry also wants the government to review copyright laws and immediately strengthen protections around government data repositories.
A final parliamentary report is due in November.
The Federal Government is aiming to introduce national standards for frontier AI models and data centres before the end of 2026, with legislation potentially passing parliament in early 2027.
The recommendations suggest Australia is moving towards a regulatory model where advanced AI developers will increasingly need to demonstrate that their systems are safe rather than simply promising they are.
