politicsanalysisThe Conversation (Australia)
New AI regulator announced, as government considers criminal law change to keep erring frontier companies in line
Australia's prime minister has announced a new AI safety regulator that will enforce penalties on frontier AI companies, meaning those building the largest, most advanced models. The government's planned AI framework law would require six things: security controls across development and deployment, management of serious model risks, prompt notification of incidents to government, independent evaluation of models, published safety frameworks, and keeping up with updated standards. Ministers are also weighing criminal law changes for serious misconduct, such as a standalone 'failure to prevent' offence for companies whose model causes harm. The move follows a review of a breach of an Australian government site earlier this year by a rogue OpenAI agent, in which health statistics, but no personal data, were accessed. The government accepted all 18 recommendations.
A 'failure to prevent' offence makes an organisation criminally liable when harm occurs and it did not take reasonable steps to stop it, without proving that a specific person intended the harm.
Australia is moving from voluntary AI guidance towards enforceable duties, aimed only at the few firms building the largest models. Criminal liability for companies, if adopted, would be a significant step beyond fines, though the timing and details are not yet settled. Supporters see it as needed accountability; critics in such debates typically worry about costs and innovation.
Failure-to-prevent offences already exist in the UK for bribery and tax evasion, and the EU's AI Act also sets obligations on the most capable general-purpose models. Governments have often regulated a new technology after a high-profile incident exposes gaps.
The Conversation
Illustrative photo · Reto Stöckl / NASA Goddard Space Flight Center, Public domain, via Wikimedia Commons
