Andrew L. Urban.
Prosecutors should not always need an AI agent to identify weaknesses in a circumstantial case. For instance, you don’t need to have been a trainee lawyer to have recognised that the police brief of the Sue Neill-Fraser case was chronically weak – too weak to take to court. I call the late, legendary defence barrister Chester Porter QC in support of that view. “There are very substantial doubts about this case … the evidence is so weak …. it is hard to see how any conviction could stand … Tasmania’s legal system risks being a laughing stock, if it wasn’t such a tragedy…” Porter said in 2013, after looking at the case of Sue Neill-Fraser – before AI. Yet then Tasmanian DPP Tim Ellis SC* in 2010 had chosen the ‘laughing stock’ option and stuck to it through the first two appeals.*
In the above example, I believe a properly constituted AI Brief Audit system would have saved Ellis from himself and Neill-Fraser from Ellis.
Circumstantial murder cases pose the greatest risk of a wrongful conviction as most of our case histories show. That’s not surprising, given the absence of direct evidence. A legally trained AI agent, such as LIA Light or LIA Pro, can provide a structured pre-trial fairness review to map what the file actually establishes, separate allegation from inference, force disclosure and further inquiry on gaps that could produce a wrongful conviction.
We have previously tasked LIA to examine several cases in our records for post-conviction assessment (eg Sue Neill-Fraser, Robert Xie, Steven Fennell, Robert Farquharson, Marco Rusterholz) and in each case serious errors or risks were identified that should have been identified in the prosecution case before the case went to trial, confirming our belief that these are wrongful convictions. The assessments were rapidly made, based on documents we supplied, such as trial and appeal transcripts, judgements, etc.
For example, in the circumstantial murder case against Rusterholz, three witnesses claimed that Rusterholz had made admissions of guilt to them; those claims became the basis of the guilty verdict and the reason for his appeal being dismissed. The AI agent found that the claims were unsupported by scientific evidence. The Appeal Court accepted the circumstantial case was not strong enough alone. Nonetheless the court held that the admissions made the verdict “safe and reasonably open”. That’s the sort of unsupportable appeal decision that keeps me busy at wrongfulconvictionsreprt.org …
affordable pre-trial brief audit
The relevant reviews would have taken considerable time if performed by humans – and hence expensive. Perhaps too expensive for any human applicant, especially after the costs already incurred and from prison.
But now, prosecutors can use LIA Light or LIA Pro before committal or trial to audit their brief at little or no cost. For best results, feed the actual brief: charges, elements, each witness account, exhibits, expert opinions, and known unused material. Require a calibrated holding on whether the elements are capable of proof beyond reasonable doubt on the present file, with each load-bearing proposition labelled evidence, allegation, inference, assumption, gap, or contradicted.
Reconstruction should apply only live issues: identity, act, intent or knowledge, causation where charged, and any defence fairly raised. The counter-reading must state the strongest innocent or alternative account in the language of these facts. Decision readiness should stay not ready wherever a gap, contradiction, or untested expert method could leave a reasonable doubt or an undisclosed line. The prosecutor then converts labels into actions: disclose, obtain, test, narrow the charge, or discontinue. That sequence reduces wrongful-conviction risk by making unsupported leaps visible before a jury is asked to make them.
On October 22, 2026, we asked the offices of the DPPs in NSW, Queensland and Victoria: has this process been contemplated, used or attempted – and would it be considered if not? Two days later, Jacqueline Harris, Media Advisor and Principal Solicitor, Policy and Specialised Legal Division, as spokeswoman for Victoria’s DPP advised that “we do not wish to provide a comment.” So far, her interstate counterparts didn’t even acknowledge my request. Maybe we are too optimistic to think prosecutors would wish to put their circumstantial cases to such a test.
Of course, defence lawyers & barristers are not infallible, but no AI agent can foresee (prophesy) defence failures as a contributing factor in a wrongful conviction.
* Tim Ellis SC – in case you are wondering why Ellis was no longer Tasmania’s DPP at Sue Neill-Fraser’s third (2021) appeal … from Wikipedia:
On 24 March 2013, Ellis was driving south on the Midland Highway in Tasmania when his vehicle crossed into a lane of oncoming traffic lane for some 30 seconds, over 700 metres. He collided with a vehicle being driven by Natalia Pearn. Ms Pearn died as a result of the crash. Ellis was found guilty of negligent driving causing death on 26 June 2014. On 15 January 2015, the Premier Will Hodgeman and Attorney-General Vanessa Goodwin announced Ellis would not continue in the role of DPP, after the Governor accepted their recommendation he be removed from the position.
I offended in a room that didn’t exist 1st . until after the event then. I offended at a time when the victim was away with her father. 2 ladies met at a time when they previously had admitted they did not know each other. And 45 other things that did not stand up to scrutiny. AI can review my petition to AG if you like. You know ,! The one that sat on a desk and was refused after 3snd a half years with no explanation or comment.” We are not obliged to tell you why you’d petition failed. “ they said. That’s Justice. ? Bring on AI. AI could come up with the correct verdict or prepare a No Bill application. Someone please give me a break. I was once a proud Australian. !!
Yes, AI may well revolutionise the legal system – in favour of the innocent.
Andrew – I think there is also an important question of where the onus for this should sit.
If the starting point is the presumption of innocence, there is a strong argument that structured review should begin with the prosecution — before the pre-trial conference.
The prosecution carries a double responsibility.
One is to the accused and the justice system: before asking a person to answer a serious criminal charge, has the State properly tested whether the evidence supports each necessary proposition, where evidence becomes inference, what assumptions are carrying weight, and what reasonable competing explanations remain?
The other is to taxpayers and society. A prosecution that should not proceed can consume enormous public resources through trials, experts, appeals, Legal Aid, incarceration and potentially years of further litigation — quite apart from the human consequences.
AI potentially makes that scrutiny much cheaper.
So perhaps the principle is straightforward: before requiring the accused to answer the State’s case, require the State to demonstrate that it has adequately challenged its own case.
That protects the accused. But it also protects the justice system and the public that pays for it.
Absolutely. Exactly why I think this is a really important question and why I am disappointed that none of the three ODPPs has engaged with the issue…perhaps naive of me?
Andrew — perhaps not naive. It may simply expose the next question.
If this kind of scrutiny can be performed cheaply, before substantial public resources are committed, what is the argument against testing it?
It doesn’t require the prosecution to surrender its judgment or authority. It gives prosecutors another mechanism for challenging their own case before asking the accused — and ultimately the public — to bear the consequences.
Perhaps the interesting experiment is therefore not whether AI can replace prosecutorial judgment, but whether it can make rigorous self-challenge cheap enough that failing to do it becomes increasingly difficult to justify.
Yes, that’s the point. I suspect that is a scary proposition for prosecutors.