Clarity for AI buildersSponsored by AI Routing
availability

Australia proposes independent testing for frontier AI models

Australia proposes independent testing for frontier AI models
Australia would require independent testing of frontier AI models under proposed rules, The Decoder reports. The proposal targets companies operating in Australia that develop the most advanced, highest-risk AI models. According to The Decoder, those companies would have to give independent auditors access to their models, conduct ongoing safety checks, publish their safety frameworks and report security incidents to the government immediately. The requirements would give regulators several ways to assess safety. Independent audits would provide a check on developers’ own assessments, while ongoing checks and immediate incident reporting would cover risks that emerge after an audit. The Decoder reports that incident reporting is intended in part to flag early signs of unexpected behavior by AI agents. A new oversight body would work alongside the Australian AI Safety Institute to enforce the rules, with penalties for violations. Prime Minister Anthony Albanese and other ministers said the government wants to put domestic safeguards in place before entering international negotiations. The Decoder reports that more details will follow after consultations; its account does not specify how models would qualify for oversight or how large the penalties would be.