Andrew Charlton stood at UTS earlier this month and told Australian businesses they were standing at a sliding doors moment .
Buy Australian AI, or become a permanent renter of intelligence from abroad. He used the economist Justin Wolfers’ thought experiment to make the point:
Imagine a worker, an employer, and an AI that does the worker’s job. If the worker owns the AI, productivity rises and everyone wins. If the employer owns it, the gains are split. If a foreign company owns it with monopoly power, the worker loses the job, the employer is barely better off, and the value flows offshore.
Three days later, Alastair MacGibbon wrote in the AFR that the rulebook every Australian board has been operating on is broken. Anthropic’s Mythos and OpenAI’s GPT-5.5 can now autonomously chain vulnerabilities a human expert would take 20 hours to exploit, and they can do it for a few dollars in tokens.
The UK government’s AI Security Institute confirmed it. Within a fortnight of Mythos being announced, Anthropic was investigating reports that unauthorised users had already accessed it through a third-party portal. The containment failed at the starting gate.
Australia was not in the coalition that got access first. Australia is, in MacGibbon’s words, not well-defended.
Get the best of Startup Daily straight to your inbox
Want to know the latest in startup news? Subscribe to our daily news and analysis coverage on what’s happening to ANZ startups, investors and the broader ecosystem. And best of all, it’s FREE!
By continuing, you agree to our Terms & Conditions and Privacy Policy .
The argument they are circling
Charlton wants the private sector to act. MacGibbon wants government to act. Both are right. Neither is enough.
The argument they are both circling without quite naming is institutional. Everyone keeps calling AI infrastructure . The Microsoft commitment last month was framed as infrastructure investment. The data centre frameworks Charlton’s department published in March were written for infrastructure. The Treasury convenes bank CEOs over AI cyber risk because the threat is now systemic. Charlton himself called intelligence the new oil.
Australia knows how to regulate infrastructure. Telecommunications has the Telecommunications Industry Ombudsman, statutory, mandatory, enforceable, created in 1993 because a deregulating market needed a dispute resolution body that could actually compel things. It has the ACCC for pricing. It has ACMA for compliance. Energy has the Australian Energy Regulator. Water has utility frameworks. When Optus went down for fourteen hours in 2023, it triggered a Senate inquiry, a regulatory review, and reporting obligations that did not previously exist.
When Microsoft’s Azure OpenAI endpoint went down for twenty hours across seven regions in early March, including Australia East, major Australian enterprise customers received zero communication. No real-time alert, no status update, no acknowledgement. The outage was detected, fixed quietly, and never publicly addressed.
There is no AI ombudsman. There is no AI pricing regulator. There are no uptime mandates. There are no outage reporting obligations. There is no statutory body any Australian customer can complain to when a foreign lab shuts down their accounts without explanation, or quietly degrades the model they pay for, or raises prices once the workflow is embedded.
The Commonwealth has done genuinely useful work on procurement, and Charlton announced more of it overnight. The rules now require agencies to approach Australian businesses first below the Free Trade Agreement threshold. The streamlined procurement threshold has been lifted to $500,000. Procurements valued at $1 million and above must consider economic benefit to the Australian economy as part of the value-for-money assessment. The Buy Australia Plan is being augmented to prioritise local AI vendors. Tech bosses from the country’s biggest businesses are being convened this month to make the case directly. These are real levers. They show what Australian government can do when it decides AI procurement matters.
What there is not, yet, is the regulatory architecture that any other piece of national infrastructure already has. The Australian AI Safety Institute was announced in November 2025 with $29.9 million in funding. Supposed to be operational early 2026. Advisory only, no enforcement power. As of MacGibbon’s piece on Monday, it still has not publicly stood up.
This is the gap New Dialogue is building to bridge. Not by replacing the institutions that should exist, but by giving Australian organisations a way to stay sovereign over their data and platform-agnostic about what runs on top of it, while we wait for the regulation to catch up.
The mistake we already made
Australia spent fifteen years treating social media as consumer software. The eSafety Commissioner came in 2015. The Online Safety Act in 2021. The under-16 ban in 2024. Each was a retrofit, after the externalities had already shaped public life, after the leverage had already moved offshore, after the pattern was visible to everyone.
The pattern was always the same. Build the infrastructure first, regulate it once the harm is undeniable. Treat the thing as a product when it suits the providers. Treat it as infrastructure when it suits the talking points. Avoid building the institutions that would actually do the regulating until the politics force the issue.
We are running the same play with AI, faster. Charlton’s procurement push asks the private sector to lean Australian. MacGibbon’s resilience push asks critical infrastructure operators to harden defences and government to negotiate access to frontier capability. Both are necessary. Both are private sector workarounds for a public failure.
If AI is infrastructure, regulate it like infrastructure. Create the body that can compel uptime reporting. Build the dispute resolution scheme any Australian organisation can use when a foreign lab makes a unilateral decision that breaks their workflow. Give the Safety Institute teeth. Stand it up. Hire the director. Publish the rules.
The moment the conversation shifts from “should businesses lean Australian” to “what infrastructure regulation does AI actually require” is the moment Australia stops carrying the load alone.
Bill Withers wrote Lean on Me in 1972, three years after he stopped working in a factory making toilet seats for 747s. He once told an interviewer the song was about the small West Virginia town he grew up in, where if your house caught fire the neighbours showed up with buckets, because nobody was coming if they didn’t.
The song is not about feeling supported. It is about who actually shows up when your house is on fire.
We have spent two months watching Australia announce frameworks, sign memoranda, host visiting CEOs, and publish strategic plans. The Safety Institute was funded in December and is supposed to be operational early this year. The plan was published. The procurement rules were updated. The speeches were given.
Where are the neighbours with those buckets?
This post first appeared on Matt Vitale’s LinkedIn .
Matt Vitale is a lawyer, capital markets adviser, a cofounder of CSF platform Birchal and now co-founder of New Dialogue, an Australian sovereign AI platform that helps enterprises deploy AI privately, securely, and without vendor lock-in. He was named Australian Fintech Leader of the Year in 2024, and writes at the intersection of technology, leadership and culture, exploring how emerging systems like AI are reshaping identity, influence and work.
Recommended for you
Australian government departments are all getting ‘chief AI officers’
Tom Williams Nov 14, 2025
Government wants Australian data centre developers to focus on renewables and startups
Poppy Johnston Mar 23, 2026
South Australian startup WorkSafe Guardian acquired by UK risk management software firm Ideagen
Simon Thomsen Oct 6, 2025


