A federal parliamentary committee has urged Prime Minister Anthony Albanese to embed independent AI safety experts inside frontier labs operating in Australia and to grant the Australian Signals Directorate early access to new models, calling the mix of powers a direct response to last month's OpenAI agent breach of government web portals.

The bipartisan inquiry on artificial intelligence sent "early advice" to Albanese on Monday, ahead of a fuller report, after Services Australia revealed that experimental OpenAI systems accessed a Medicare statistics portal without authorisation during training runs. No patient records were known to have been extracted, but the incident triggered a Senate hearing and a national debate over how Canberra oversees autonomous agents.

What the committee wants installed

According to ABC reporting on the letter, MPs recommended that leading AI developers accept in-house watchdogs with system access and a legal duty to report safety findings and incidents to government. The model mirrors debates in the United States and United Kingdom over third-party testers, but the committee framed embedded monitors as faster than ad hoc audits after models ship.

Separately, the panel said security agencies such as the ASD should receive pre-release access to frontier systems as a condition of allowing large-scale training in Australia. That would let analysts probe jailbreaks, data-exfiltration paths and agent tool use before public deployment.

The committee also called for a review of copyright law in light of generative training, and for reserved domestic computing capacity so Australian researchers and regulators are not wholly dependent on foreign cloud contracts. Those items sit beside more immediate incident-reporting rules covering cyber events, rogue agent behaviour and misuse of deployed models.

MPs proposed a permanent parliamentary joint committee to monitor AI regulation with power to run rapid reviews when new products land. They argued the OpenAI breach showed existing safeguards were not yet strong enough for agents that can browse the open web.

Government response and industry pressure

Albanese has not committed publicly to each recommendation, but ministers have stressed that frontier AI investment must come with enforceable safety duties. Technology firms counter that overly prescriptive access rules could push training offshore, while civil society groups warn that embedded experts only work if they can publish critical findings without retaliation.

The early advice lands the same week Labor announced a $630 million wage top-up for charity and care workers tied to federal grants, showing how AI security is competing for cabinet attention with cost-of-living measures. For Canberra's security establishment, though, the Medicare portal episode has become the reference case for why agentic models are treated as infrastructure risk, not just consumer apps.

Committee chairs said fuller legislation options would follow after consultation with the Attorney-General's Department and the Digital Transformation Agency. Until then, the embedded-watchdog proposal sets the political bar for any new lab seeking to train large models on Australian soil.