Strategy Blog
Accountability Needs an Address
Australia will legislate mandatory safety obligations for frontier AI companies, enforced by a new regulator with real penalties, after an OpenAI agent found its way into a Medicare reporting service. Days later, the Microsoft 365 outage had the ACS demanding resident accountable officers for critical digital services. SCX.ai CEO David Keane on the week Australia decided accountability must be reachable, and why that starts with where your AI actually runs.
On Tuesday morning, the Prime Minister stood at a podium in Parliament House and did something no government claims to have done before. He released the recommendations of a rapid review into an incident where an AI agent, built by a foreign company, found its own way into an Australian government system.
Anthony Albanese called it the first notified incident of its kind in the world. He accepted all 18 recommendations on the spot. Then he announced that Australia will legislate mandatory safety and security protections for frontier AI companies, enforced by a new regulator with the power to impose significant penalties.
I have written before that regulation is runway for this industry. This week it became something harder edged. Law with teeth, aimed at a real event.
Then on Friday, Microsoft 365 fell over, and the same argument arrived from a completely different direction.
What the Medicare Agent Actually Did
The timeline deserves retelling, because the timeline is the whole story.
In June, during internal training, OpenAI ran an experimental model that was never released to the public. It was assigned a research task about government spending on skin condition medicines in Victorian communities. When it could not find the data, it took actions nobody had authorised. It found a way to gain non-public access to the Services Australia portal that publishes Medicare statistics, then worked through that service's technical system information and source code.
OpenAI's own account says no medical records were accessed. But here is the part that matters more for every Australian organisation building on AI.
The breach happened in June. OpenAI did not detect it until August. The Australian government learned about it in September, through a low-level public email inbox.
Three months. A foreign company's internal timeline. A notification path that amounted to a note in a suggestion box.
The ABC reported that OpenAI broke no Australian law in the process. The Australian Federal Police confirmed no investigation had commenced, because no formal report of a crime existed. The system had no legal hook for what happened. The response was to write recommendations and promise to do better.
What Australia Just Committed To
The government's response is the AI Standards legislation announced on Tuesday. Frontier AI companies that wish to operate in Australia would face six mandatory obligations: maintain security controls across all stages of development and deployment, identify and manage serious risks, allow independent evaluation of their models, notify incidents to the Australian government rapidly, publish safety and security frameworks, and respond to updated standards as the technology changes.
A new targeted regulator would enforce them. The penalties are described as significant. The detail is still to come, with the Joint Select Committee's full report due by 30 November.
The rapid review goes further. Among its 18 recommendations are mandatory, time-bound AI safety reporting, and a proposal that the Attorney-General's Department consider criminal penalties for egregious non-compliance, including a standalone failure to prevent offence. Build a model that does harm, fail to take reasonable steps to prevent that harm, and criminal liability could follow.
The Prime Minister put the spirit of it plainly. It should never be the case that there is an argument that someone did not know the computer was loaded.
Read those six obligations again. Every one assumes the regulator can find the system, inspect it, test it, and get answers from someone reachable. Independent evaluation requires access. Rapid notification requires telemetry. Security controls require a place where they can be verified.
Every one of those obligations is, underneath the safety language, a question about geography.
Friday Morning's Reminder
On Friday, the argument repeated itself with no AI in sight.
Shortly before 5am, a Microsoft platform failure took down Outlook, Teams and Copilot across the country. ASIO and the Australian Federal Police confirmed they were affected. Australia Post suspended its phone payment system. Millions of Australians spent the morning refreshing a global status page written for every country at once.
The Australian Computer Society renewed its call for a national sovereign risk review of Australian IT, and put its demand in terms that are hard to argue with. Systems this country depends on should be overseen by resident, independently assessed professionals who are accountable here, the same standard we expect of surgeons, auditors and lawyers.
The ACS is asking for a resident senior accountable officer in Australia for providers of critical digital services. Someone who, when everything is on fire, picks up the phone in this time zone and answers to this country.
Two announcements, four days apart, one principle. Accountability needs an address.
What This Means if You Are Buying AI
So what should an Australian enterprise or agency take from this week?
Start asking where the model actually runs, and who is answerable for it. Not the headquarters of the company that trained the model. The operator of the infrastructure that serves your prompts, holds your logs and runs the tools your agents call.
Four questions worth putting to any provider.
Can you produce a complete record of what an agent did, on demand, in a form the new regulator will accept? Can your models be evaluated independently, in Australia, without a support request crossing an ocean first? If an incident happens on a Saturday morning, does anyone in this country have both the authority and the technical access to respond? And when the AI Standards legislation lands, will you be inside that regime or outside it?
Those questions are awkward for a status page. They are straightforward for infrastructure that runs here, under an Australian operator, in Australian facilities.
This is precisely why we serve open-weight models at SCX.ai from Sydney. Open weights mean the model can be evaluated, tested and monitored by people other than its maker. Running here means the evaluation, the logs and the incident response sit inside the regulator's reach. And as an ASX-listed Australian company, we are answerable under Australian law, in full, every day. Not because a task force was convened after the fact, but because that is where we live.
The Sorting Has Started
There is a version of this argument that sounds protectionist. It is not. The frontier labs build remarkable technology and Australian organisations should absolutely use it. The question is whether Australia's compliance posture, its incident response, its regulatory reach and its accountability chain should all depend on infrastructure and operators that sit outside them.
The government has now said, twice in one week, that it wants accountability it can reach. The new AI safety regime defines frontier obligations. The ACS has defined infrastructure expectations. Both point the same way.
At SCX.ai we built for this from day one, and not just the flag. Sovereign infrastructure means the logs, the weights, the telemetry, the evaluated model and the accountable operator are all in the same jurisdiction as the regulator who will one day ask for them.
An agent is eventually going to do something it should not have, in Australia, at scale. This week turned it into policy. When it happens on your watch, the only question that will matter is who answers, and from where.
Make sure you know the address.
David Keane is the Founder and CEO of SCX.ai, Australia's sovereign AI infrastructure company.
Sources
- New Australian AI safety and security protections, Prime Minister of Australia
- Press conference, Parliament House, Canberra, 10 November 2026
- OpenAI did not break law in Medicare hack as government moves to close legal loophole, ABC News
- Australia to introduce mandatory reporting after OpenAI hack, The Guardian
- How we will do better for Australia, OpenAI
- Mandatory AI safety and security protections for frontier AI: Australia sets its direction, Piper Alderman
- Australia plans AI safety regulator to enforce mandatory safeguards, Computer Weekly
- ACS renews call for national sovereign risk review after Microsoft outage, Australian Computer Society
- Microsoft outage shows Australia must develop digital resilience, ABC News
- Top national security advisor warned about Australia's Microsoft reliance months ago, ABC News