Australia's parliamentary AI committee wants independent evaluators made mandatory inside frontier AI labs. In an October 12 letter to Prime Minister Anthony Albanese, chair Jo Briskey and deputy chair Jonathon Duniam propose giving them employee-like access to systems and operations.
The outside experts would report safety tests and incidents to government under established guidelines and timeframes. Labs would also have to monitor risks and report incidents against clear deadlines. The letter does not specify those deadlines or how evaluators would be selected and funded.
A permanent joint parliamentary committee would oversee the industry and publish details of unauthorized AI-agent actions. For arrangements allowing frontier-model training in Australia, the letter also recommends early model access for agencies such as the Australian Signals Directorate.
The process drew criticism. Senator David Pocock told ABC he objected to the prime minister's office reviewing the draft beforehand. Briskey defended the consultations; Duniam denied interference. The recommendations are early advice from the committee, which says its full report is due by November 30.