JENI O’DOWD: Anthony Albanese’s AI policy was recently announced, but it’s missing something

There’s a bloke I once worked with who thinks AI can solve just about anything, from writing his emails and reports to practically telling him what to think. What he doesn’t realise is that every email or report he gets AI to write for him is glaringly obvious.

I’m happy to give AI the boring stuff. But it will never replace human judgement. The Federal Government, it turns out, understands that when it comes to creatives.

But it seems far less interested in everyone else.

A few weeks ago the Government released a policy on protecting Australians from AI, which sets rules for the data centres that power AI, and promises Australian writers, artists and journalists will keep ownership of their work.

It’s a genuine protection, and a great one. But what it missed entirely was the danger of companies using AI to do their own work.

There’s nothing in it about accountability, disclosure, or checking before you sell work done by AI to someone else.

Here are some examples.

In July last year, Deloitte handed the Department of Employment and Workplace Relations a 237-page report on the Government’s welfare penalty system.

Taxpayers paid $440,000 for it. It took a university researcher, Chris Rudge, to actually check the footnotes and find a quote from a Federal Court judge that the judge (whose name was misspelt) never said, and citations to academic papers that don’t exist.

Deloitte had used AI to write parts of the report. It seemed nobody had bothered to check its work before sending the invoice. The firm eventually refunded $97,000 and admitted AI had been used to help write it, but kept every recommendation in the report exactly as it was.

Deloitte wasn’t alone. Global firm KPMG pulled its own report on the future of agentic AI this year after discovering only five of its 45 citations were real.

Many major organisations, including UBS, the UK’s National Health Service, Swiss Federal Railways and Transport for London said the claims it made about their AI use were simply wrong.

EY Canada withdrew one too, a cybersecurity report used in client briefings, built on a McKinsey statistic about a “$200 billion loyalty market” that doesn’t exist anywhere in McKinsey’s archives.

So three of the biggest consulting firms in the world have all been caught handing over invented facts dressed up as expert advice, with all of it happening while this Government was busy writing an AI policy about something else entirely.

Deloitte and EY both called it an “error” or a “process failure”, but somewhere down the line, someone signed off on the reports and sent off an invoice.

The Department of Employment and Workplace Relations was meant to use the Deloitte report to help shape decisions about how vulnerable jobseekers get penalised.

Real people’s welfare payments, run through a system partly justified by a court quote that was never said and a book that doesn’t exist.

That’s not a technical glitch. That’s a consulting firm charging taxpayers nearly half a million dollars for some work it never actually did.

Back to the Government’s recent AI policy, which Prime Minister Anthony Albanese described as a “world-leading framework” about “Australia choosing to shape the future rather than letting the future of AI shape us”.

“This framework is about protecting our national interests and ensuring certainty for growth, jobs and investment,” he said.

“If we set our national standards high, then we can make AI stand for Australia’s interests.”

It seems he missed the memo on potential fraud.

It’s not like the Government can’t move quickly on AI. It did exactly that in 2024, making deepfake sexual material a Federal crime with a penalty of up to seven years imprisonment.

That’s the right call. But there’s still no equivalent seriousness for a firm that hands the government or a client a partially fabricated report and gets paid for it, or gets business from it.

And most importantly, there is no acknowledgement that this is probably happening in offices all over the country.

Some bosses are asking staff to use AI to do more with less. Deadlines are tighter. Teams are smaller. The pressure to deliver is enormous. So people use AI to fill the gaps, hit the deadline and get the thing done.

Nobody has set clear standards. Nobody has built in proper checking. Nobody has demanded disclosures. And so, slowly and quietly, fabricated content is being laundered as professional work.

Deloitte, KPMG and EY were caught because someone was paying attention. But most of the time, nobody is.

The Government does have an AI Safety Institute, launched in December with $29.9 million in funding (not surprisingly, a few months after the Deloitte case made headlines).

The institute tests frontier AI models and writes voluntary safety standards. In other words: please police yourselves.

We all know how that’s going.

Treasurer Jim Chalmers has knocked back union calls for AI regulation in the workplace, saying he’s focused on “capabilities and opportunities, not just guardrails”.

None of this makes AI the villain, by the way. Used properly, with an actual human checking the work and real accountability behind it, it’s a genuinely good tool.

But the words “used properly” are doing a lot of heavy lifting, and right now nobody is defining what that actually means.

So we are left with a culture where the pressure is to produce, the incentive is speed, the oversight is minimal, and the consequences for getting caught are, at worst, a partial refund and a quiet apology.

In that environment, cutting corners is not a scandal. It is a business model.