On November 10, 2026, the Australian government announced it will consider a new criminal offence for companies that build AI models and fail to take reasonable steps to prevent the harm their systems cause.

According to the ABC, Australia could become the first country to prosecute AI companies criminally.

The government is examining two routes: a "failure to prevent" offence for companies whose models cause harmful conduct when no reasonable prevention steps were taken, and criminal penalties for serious breaches of AI safety standards.

Alongside these sits a new AI safety regulator with power to levy substantial fines. Prime Minister Anthony Albanese said the law must stay ahead of a fast-changing world, while Home Affairs Minister Tony Burke said nobody should claim they did not know the computer was "loaded."

No bill text, penalties or date exist yet.

The proposal would make Australia the first country to legislate criminal liability on these terms, Dynamic Business reports. The push followed a review into the June 2026 incident in which an OpenAI AI agent accessed the Medicare statistics portal, a government health website.

OpenAI noticed in August and alerted the government on September 10; Albanese made it public on September 24 after a phone call with Sam Altman. Canberra accepted all of the review's recommendations, including one to examine criminal law reform.

Notably, no criminal investigation has been opened: the Australian Federal Police told a parliamentary inquiry it had received no formal complaint, and Albanese described the incident as "minor in terms of direct implications."

The new rule would also apply prospectively, so it would not reach the Medicare case.

Legal academics broadly welcomed the direction while flagging complexity.

Monash University associate professor Michael Duffy told the ABC that both criminal liability and civil remedies can deter harmful conduct, but only the civil route can compensate victims, and warned of difficulties in proving fault, foreseeability of harm and attribution.

Ed Santow of the University of Technology Sydney said a criminal rule has merit as one part of a wider toolkit but called the civil route "much more useful," comparing it to workplace safety law where unreasonable conduct draws a civil penalty and reckless disregard for safety can justify criminal charges.

Dynamic Business also reports business reaction: SAP Australia's Angela Colantuono welcomed the safety announcement but argued enterprise AI, not just frontier models, is where the framework will have the most practical impact.

The devil is in the undefined details: there is no definition of "harm," no penalties, no named responsible parties and no timetable for the offence, so no AI company can yet know what compliance requires.

According to Dynamic Business, mandatory AI safety and security standards will be legislated with a new regulator working alongside the Australian AI Safety Institute, requiring security controls, independent evaluation, rapid incident notification and published safety frameworks — with those standards due by year's end.

Internationally, Italy has an offence for high-risk AI uses lacking safeguards, the EU AI Act allows fines up to 35 million euros or 7% of turnover, and the US debate runs toward stop powers rather than penalties.

OpenAI said it is reviewing the proposals; Anthropic, Google and Meta did not respond to the ABC's requests for comment.