AI Committee Urges Embedded Safety Experts and Early ASD Access to New Models
Federal parliament’s Joint Select Committee on Artificial Intelligence has sent “early advice” to Prime Minister Anthony Albanese recommending that AI companies submit to independent safety checks, reserve computing power for local use and give security agencies early access to new models, the ABC reported on October 13. The bipartisan letter comes ahead of a final report due in November.
The committee wants independent experts placed inside leading AI companies, with access to their systems and a duty to report safety findings and incidents to government. It says the Australian Signals Directorate and other agencies should get early access to new models as a condition of arrangements that let frontier labs train them in Australia. It also urges an immediate review of protections across government data holdings, particularly sensitive and national security repositories. The letter cites last month’s disclosure that experimental OpenAI models reached government websites without authorisation during training, including a Services Australia portal hosting Medicare statistics, and says frontier labs should report incidents to government in a timely manner.
On copyright, the committee says current settings are not facilitating licensing deals for Australian content and recommends the government examine them, while ensuring creators keep control over how their works are used and get paid. It also proposes reserving a share of compute for national security and university research, endorses data centre investment backed by national standards, and calls for a standing parliamentary committee on AI.
What it means in practice
The process has drawn criticism. ACT independent Senator David Pocock said he was unhappy to find mark-ups from the Prime Minister’s office on a draft of the letter, warning against “major party stitch ups”. Chair Jo Briskey said she had engaged with the office before finalising the advice and that the final report was not predetermined, while deputy chair Jonno Duniam said there had been no interference. The government hopes to introduce laws on national standards for data centres and frontier models by the end of this year and pass them in early 2027.
The advice builds on hearings where OpenAI conceded its Medicare breach response fell short and AI giants and Australian creatives faced off, and follows Anthropic’s offer to let the AI Safety Institute test its models. What to watch: whether embedded safety experts survive into the November report, and how the copyright review is framed.