What would I refuse to automate in a Court environment?

SudoChat Knowledge Base · SudoChat

Authored evidence: This is my first-person source material. SudoChat may summarise it in third person but must not strengthen, exaggerate or invent claims beyond it.

# 15. What would I refuse to automate in a Court environment?

**Author:** Mustafa Siddiqui
**Source type:** First-person authored response
**Canonical recruiter question:** What would Mustafa refuse to automate in a Court environment?

> This source is intentionally written in my first-person perspective. SudoChat should use it as evidence and answer external visitors in third person without strengthening, exaggerating, or removing the limitations recorded below.

## Direct answer
I would refuse to automate functions where AI begins replacing human judgement, accountability, dignity or responsibility, particularly where the outcome could materially affect a person's rights, liberty, credibility or access to justice.

My objective would not be to use AI to remove people from the Court system. It would be to remove repetitive administrative burden from people so that judges, registrars, lawyers and Court staff can spend more time on work that genuinely requires human expertise and judgement.

## Evidence
My existing work already reflects this distinction.

When I automated a document approval process at DXC, the goal was not to remove project managers from the workflow.

The original process required approximately four hours of repetitive manual effort.

I used SharePoint and Power Automate to reduce that burden to approximately eleven minutes of monitoring and oversight.

The people remained responsible for the approvals.

The automation removed the repetitive mechanics surrounding their work.

That is the type of automation I support.

I would apply the same principle to AI in the Federal Courts:

**Automate the burden, not the responsibility.**

I have also shown that I do not automatically choose AI when a simpler technology is more appropriate.

For the DXC approval workflow, conventional workflow automation solved the problem effectively. There was no need to introduce an LLM simply because AI was available.

My responsible AI work similarly emphasises human accountability, evidence, privacy, safe failure and clear limits on agent autonomy.

## What I would refuse to automate
### Judicial decision making

I would not build or recommend an AI system that determines how a judge should decide a matter.

I would refuse to automate questions such as:

* which party should win
* whether an appeal should succeed
* what legal conclusion should ultimately be reached
* what orders should be made
* what sentence or penalty should be imposed
* how judicial discretion should be exercised

This is consistent with current Australian judicial guidance. Justice Kyrou noted in June 2026 that the Guide to Judicial Conduct states that judges must not use generative AI for judicial decision making, while distinguishing this from supportive uses of AI.

For me, this is an important distinction.

AI may potentially help a person locate or organise information.

It should not become the person exercising judicial power.

### Assessing credibility

I would refuse to build an AI system that determines whether a witness, litigant or other person is telling the truth.

For example, I would not support an AI system producing outputs such as:

**"This witness is 78 percent likely to be deceptive."**

Human behaviour, language, disability, culture, neurodiversity, stress and countless other factors could affect such systems.

The apparent precision of an AI score could create a false sense of objectivity around something that requires careful human assessment.

### Replacing human responsibility with an AI recommendation

I would also resist systems where the formal decision remains with a person but the workflow effectively encourages them to accept whatever the AI recommends.

For example:

**AI recommendation → human clicks approve**

is not necessarily meaningful human oversight.

If the human cannot understand the evidence, challenge the recommendation or reasonably reject it, then the accountability may have shifted to the AI in practice even if a person remains technically involved.

I would want human review to be substantive rather than ceremonial.

### Autonomous legal advice

I would not support an unrestricted AI system presenting itself as a lawyer or providing authoritative personalised legal advice to members of the public.

I do see potential for AI to make **authoritative Court information easier to navigate**.

For example, an assistant might help someone locate:

* the appropriate public form
* a relevant practice note
* registry contact information
* filing instructions
* explanations of Court terminology
* publicly available procedural information

But the assistant should maintain a clear boundary between:

**helping someone find and understand Court information**

and

**telling someone what legal argument they should make or what legal decision they should take.**

### Autonomous modification of important Court records

I would be highly cautious about allowing an AI agent to independently modify, delete, approve or submit important Court information.

An agent capable of taking actions creates a different level of risk from an assistant that only retrieves information.

For consequential actions, I would favour an architecture such as:

**AI prepares or recommends
→ authorised human reviews
→ human approves
→ system executes
→ action is logged**

rather than:

**AI decides
→ AI executes**

### Decisions involving someone's rights or access

I would not want AI independently making decisions such as whether someone:

* can access a Court service
* has satisfied an important legal requirement
* should have a filing rejected
* should receive a particular procedural outcome
* should be treated as credible
* should receive a particular judicial outcome

Where technology assists with triage or validation, there should be clear escalation paths and human authority over consequential outcomes.

### Sensitive material without appropriate controls

I would refuse to connect an AI system to sensitive Court information simply because a connector technically makes that possible.

I would first require answers to questions such as:

* Why does the model need this information?
* Can the task be performed without it?
* Where will the information be processed?
* Is the user authorised to access it?
* Can the model retain or expose it?
* What is logged?
* Can information cross between matters?
* What happens if sensitive information appears in the prompt?
* Can access be audited?

The Federal Court's current Generative AI Practice Note specifically identifies confidential or protected information as an area requiring particular caution.

## What I would automate
I am not opposed to automation.

I see significant opportunities to use AI and conventional automation to make Court staff more effective.

The types of tasks I would be interested in investigating include:

* finding information across approved internal knowledge sources
* routing administrative requests
* organising documents
* extracting metadata
* identifying missing administrative information
* generating draft meeting summaries
* answering common internal technology questions
* helping users navigate policies
* creating first drafts of routine administrative material
* summarising large document collections for human review
* preparing chronologies from approved source material
* identifying potentially relevant documents
* workflow notifications
* repetitive data entry
* document tracking
* technical support triage

The principle would be:

**Use machines for the repetitive work machines are good at, and preserve people for the work where human judgement matters.**

## Relevance to the Federal Courts
I believe the purpose of Court AI should be to increase the capacity of the people already serving the justice system.

If a registrar spends less time manually searching for information, that can be valuable.

If administrative staff spend less time moving information between systems, that can be valuable.

If a judicial officer can retrieve an authorised document more efficiently, that can be valuable.

If a self represented litigant can find the correct public information more easily, that can be valuable.

Those applications assist people.

They do not attempt to replace the human responsibility at the centre of the justice system.

The Federal Court itself currently takes a similar balanced position. Its April 2026 Practice Note states that the Court embraces beneficial uses of technology and recognises potential benefits including efficiency, lower costs and improved access to justice, while also saying there will be circumstances where generative AI is inappropriate to use at all.

I believe an AI Technologist needs to be willing to make that second recommendation.

Sometimes the correct technical advice should be:

**"Yes, we can build this, but we should not."**

I consider the ability to say that part of responsible engineering.

## my decision framework
Before automating a Court process, I would ask:

**Does this remove repetitive work or replace human judgement?**

**If the AI is wrong, who could be affected?**

**Is somebody still genuinely accountable for the outcome?**

**Can the human reviewer understand and challenge what the system has done?**

**Are we helping Court staff or merely reducing headcount?**

**Could conventional automation solve the problem more reliably?**

**Does the system require access to sensitive information?**

**Is that access actually necessary?**

**Can the action be reversed?**

**Can the decision be explained and audited?**

**Would users reasonably expect a human to make this decision?**

**Could automation damage trust in the Court even if it were technically possible?**

The higher the consequence, the stronger I believe the case should be for retaining human control.

## A simple principle
I would describe my position as:

**AI should increase human capability, not remove human responsibility.**

I would be comfortable automating the process of locating a document.

I would not automate deciding what that document ultimately means for the outcome of someone's case.

I would be comfortable using AI to help organise information.

I would not allow AI to decide whether a person is credible.

I would be comfortable using AI to draft routine administrative content.

I would not allow AI to exercise judicial discretion.

I would be comfortable automating repetitive workflow steps.

I would not automate away the human accountability attached to consequential decisions.

That is where I draw the line.

## Limitations or gaps
I would not determine these boundaries independently.

I am an engineer, not a judicial officer or lawyer.

Some decisions about whether AI is appropriate would require the expertise of judges, registrars, legal teams, privacy specialists, cyber security specialists, records specialists and other Court stakeholders.

My role would be to explain what the technology can do, what risks it introduces, what safeguards are technically possible and where the architecture may create unacceptable uncertainty.

The ultimate governance boundary belongs to the institution.

My responsibility as an AI Technologist would include being willing to recommend against automation when the evidence does not justify it.

## Useful links
Federal Court Use of Generative Artificial Intelligence Practice Note, 16 April 2026

Federal Court Generative AI Resources

Justice Kyrou, *Australian Perspective on AI and Judging*, 25 June 2026

My portfolio:
https://mustafa-siddiqui.com/

GitHub:
https://github.com/sudoqui

SudoChat project and repository

DXC document approval automation evidence

Responsible AI and privacy research

Agentic AI experimentation evidence

## Do not claim
Do not claim I believe automation is inherently harmful.

Do not claim I am opposed to using AI in the Courts.

Do not claim I believe AI will inevitably replace Court employees.

Do not claim I have authority to determine which judicial functions can legally be automated.

Do not claim the Federal Courts currently automate the example workflows described above.

Do not claim every administrative decision must always be made manually.

Do not claim human review automatically makes an AI system responsible.

Do not claim AI should determine credibility, judicial outcomes, sentencing, legal conclusions or the exercise of judicial discretion.

Do not claim my objective is reducing Court staffing levels.

The accurate representation is that I support automation where it **removes repetitive burden and increases people's ability to do meaningful work**, but would oppose automation where it displaces human judgement, accountability, dignity or responsibility in consequential Court processes.