Senate debates

Wednesday, 8 November 2023

Answers to Questions on Notice

Question No. 1917

3:37 pm

Photo of David ShoebridgeDavid Shoebridge (NSW, Australian Greens) Share this | Hansard source

I think that was an explanation, and therefore I move:

That the Senate take note of the explanation.

Not only did we email the minister's office—and of course there seems to be a breakdown in internal communications there—but this question was asked on 11 April. They've had some seven months to provide an answer to a question which, I've got to say, other ministries and other ministers were able to answer within four weeks. The questions we were asking of the Treasurer were: are the departments, agencies or ministerial offices under the Treasurer's direction using ChatGPT? It is a pretty simple question. And what did we get? Silence. We said: if yes, what were they using it for? No answer. Then we said: which bits of the agencies are using ChatGPT? No answer. We asked: do the ministers have any policies prohibiting employees? No answer. We asked: can you explain, if it is used, what the rationale for the use is? Why can't the Treasurer answer this within seven months? Why can't the Treasurer say whether or not ChatGPT is being used in Treasury and other agencies under the Treasurer's control?

This is an important matter of public interest because it fundamentally relates to privacy, to security and also to data sovereignty. We know, from ministers who actually take their accountability to parliament seriously, that multiple departments have been using ChatGPT, and we know they've been doing it without any policies in place, without any record keeping and without any checks and balances being put in place. How do we know that? Because those other ministers actually answered questions directed to them from parliament. They actually read the rulebook on Westminster accountability.

For example, we know that Department of Defence staff accessed ChatGPT thousands of times without any approval. It seems that the Department of Defence went off on a frolic on ChatGPT, maybe hoping that was the way they could solve some of their seemingly insoluble problems about recruitment and procurement or even just try to work out how many generals, air marshals and admirals they have at any one point in time. It probably would require a detailed request of ChatGPT to actually track down the number of senior personnel in the ADF. But, anyhow, they managed to respond to the questions put to them.

We also know that Home Affairs have been experimenting with ChatGPT. They've been using it in the Information and Communications Technology Division, in the Refugee, Humanitarian and Settlement Division, in the Data and Economic Analysis Centre and in the Cyber and Critical Technology Coordination Centre. Remarkably, those parts of Home Affairs whose job it is to do cybersecurity and to look at critical technology coordination had all been plugging away into ChatGPT for experimentation and learning purposes with no record keeping and no policy. Heaven knows how much of our public and private data Home Affairs fed into ChatGPT. The reason there's a big question mark about what Home Affairs did is that they kept no records at all of their use of ChatGPT.

The only way they were able to try and come up with at least some kind of answer about the use of ChatGPT was by putting a survey out to their staff, saying: 'Please tell us if you've been using ChatGPT. Tell us if you can remember what it was for. Tell us if you can remember when it was done. Tell us if you can remember the nature of the response.' A bunch of the responses, when we looked at the survey, which we managed to force out of them, were: 'Look, I used it, but it was a while ago. I can't really remember what I plugged in. Maybe I should be able to remember what I plugged in. Sorry about that.' That's Home Affairs, which is meant to be leading our cybersecurity response. Not only did they have no policy, and not only did they keep no records, but, when they surveyed their staff, they couldn't even remember.

So is it any wonder we want an answer from Treasury about what they've been doing with ChatGPT? Is it any wonder that it's not just the Greens who think that the government should have transparency on how they use these large language models, with all of their potential biases, privacy concerns and security concerns? Is it any wonder that we want an answer to these questions? And we want an answer within 12 months. It's been more than seven months since we asked the question—seven months waiting for an answer from the Treasurer about the use of ChatGPT. And then, when we put them on notice today, the Special Minister of State seemed to be lost in some sort of email black hole. They can't even get notification from Treasury about the request that we made and finally provide an answer.

Why is it important that we get the answer from Treasury? One of the key reasons that Treasury needs to come clean about its use of ChatGPT is that it has so much access to sensitive and confidential information. It has detailed predictions about where the economy is going to go. It's making hugely important decisions about the allocation of public finances. It's preparing the national budget. So one of the things we want to know is: did the Albanese government check on ChatGPT for the preparation of their last budget? Is the Albanese government currently feeding information into ChatGPT to come up with speeches or to come up with other public policy positions in relation to the upcoming budget? Are there any policies at all in Treasury controlling how they use ChatGPT?

I've got to say there's a deafening silence from Treasury, and behind the deafening silence you can actually hear the alarm bells going off. The real concern with the Treasurer refusing to answer the questions about ChatGPT is that they've been using it a lot without records, without policy and without any checks and balances. How else do you explain a seven-month embarrassed silence from the Treasurer about whether or not they use ChatGPT and the circumstances in which they use it?

We've seen Treasury behaving very badly with public information already. We've seen how few controls they have on critical public information. For example, remember how they gave a private briefing to PwC and a bunch of other leading global accountancy firms? They gave a private briefing about proposed changes they were making to Australia's tax laws, and they gave that detailed heads-up—critically important and confidential information, information that if improperly disclosed could cost the Australian public and Treasury hundreds and hundreds of millions of dollars. But Treasury were okay handing that out to PwC, if you remember. They handed it over to PwC for whatever. Heaven knows what was going through their minds at the time—that they thought it was a good idea to hand out that confidential information to firms whose job it was to advise corporate Australia on how to avoid tax. Heaven knows how they thought that was a good idea to give that information to PwC.

The question is: have they been equally reckless with the information that they've been feeding into ChatGPT? Once you feed public data or a citizen's data into ChatGPT, it's gobbled up by the model. It's literally swallowed by the model, and you can never get it back again. So how much public information, public data and confidential information has Treasury fed into ChatGPT? Well, I'd like to know. I think the public has a right to know. The Treasurer should answer the question, and the minister responsible here should make the Treasurer answer the question because that's how the system's meant to work. It is so unreasonable to wait seven months to get some basic transparency out of Treasury.

We also know that if that Treasury information is fed into ChatGPT then it can be pulled out and extracted by other people. If that information gets into the wrong hands, it can obviously cause harm, and it's vital therefore to know what data they've entered into products such as ChatGPT, with all the serious privacy and security concerns that that holds. It is more than frustrating to see a government that came in with all of that baloney about transparency and those statements about saying it'd be different to the previous government; this time they would be transparent and this time they'd give information and wouldn't be the same as the secret coalition government under Morrison. This was going to be a new government, and bloody hell we needed a new government. But it is so frustrating to read all those promises from when Labor was in opposition and then to see Labor in government. It's like the leopard changed its spots as soon as it got into the government benches.

We had the Attorney-General, when he was in opposition as the shadow AG, supporting challenges in the Federal Court to the failure to provide answers to FOI, but as soon as he became the AG he funded the opposition to that very same case to the tune of hundreds and hundreds of thousands of dollars to shut down information. And we had the Labor opposition complain about the lack of answers and transparency from government ministers, and now they're sitting there in the Treasury benches, in the government benches, and behaving just like the coalition did when it came to transparency.

All that talk and all that bluster, and here we are seven months down the track still not having a basic answer about the extent to which Treasury used ChatGPT. That's a shameful indictment of the lack of transparency from the Albanese government, and it puts into question all of those other public commitments they made.

So I ask again, not just on behalf of the Greens but on behalf of the Australian public: what's Treasury been doing with ChatGPT? How much of the Australian public's data have they fed into it? What checks and balances did they have on it? What, if any, policies do they have in place to protect this confidential information, which is critical to the finances of the federal government and critical as well for the private protection of citizens' personal and private data. It is not too much to expect an answer. An answer within 28 days is what's required, but here we are seven months down the track, and we have this deafening silence from Labor and the Treasurer on the extent to which they've used ChatGPT. I would expect that, in the absence of a rapid answer from the Treasurer, we'll be back again sometime later this week demanding some transparency and some adequate answers from the Treasurer. If you've got nothing to hide, Treasurer, give us the answers.

Question agreed to.

Comments

No comments