AI-Assisted Document Review After Australia’s New Federal Court Guidance

Generative AI is moving quickly into document review, investigations and litigation support. Legal teams can now use AI tools to summarise documents, identify themes, extract key information and help prioritise large collections. This may reduce the time required to understand a matter, especially during early case assessment or an internal investigation. However, using generative AI in litigation also creates questions about accuracy, confidentiality, privilege and responsibility for the final work product.

On 16 April 2026, the Federal Court of Australia issued a new Practice Note setting out its expectations for the use of generative artificial intelligence in court proceedings. The guidance is relevant to document review because it expressly covers pleadings, written submissions, lists of documents and other material provided to the Court.

The Court Has Not Banned Generative AI

The Federal Court recognises that generative AI may improve efficiency, reduce legal costs and support the just resolution of disputes. The Practice Note is not a prohibition on AI use. Instead, it makes clear that users must understand the technology’s capabilities, limitations and risks.

A person may also be required to tell the Court whether generative AI was used, how it was used and for what purpose. Responsibility remains with the person submitting the material. Using an AI tool does not transfer responsibility to the software provider.

Why Document Review Is Affected

The Practice Note specifically refers to lists of documents prepared for discovery.

It does not prescribe how generative AI may be used during the underlying review. However, where AI assists in preparing a list of documents, the responsible person must still confirm that the completed list complies with the Federal Court Rules.

As a practical matter, legal teams should independently verify any AI-assisted classification or description that contributes to the list.

An AI system might help identify potentially relevant records or generate initial descriptions. A lawyer must still decide whether a document is discoverable, whether privilege applies and whether the description is accurate. This is particularly important where the AI system has only been given selected documents, incomplete metadata or extracted text that does not fully represent the original file.

Generative AI Has Not Replaced Technology-Assisted Review

In an industry analysis published on 7 April 2026, MinterEllison described Continuous Active Learning, or CAL, as the current gold standard where a transparent and defensible review methodology is required.

This is a practitioner assessment rather than a finding in the Federal Court’s Practice Note. However, Australian courts have previously accepted Technology-Assisted Review and CAL as appropriate discovery methods in larger matters. CAL learns from human coding decisions and prioritises documents predicted to be relevant. Its performance can be assessed through documented validation, commonly including statistical sampling of documents predicted to be non-responsive.

Generative AI provides different advantages. It can rapidly summarise material, answer questions, extract information and help legal teams identify issues early.

A practical workflow may therefore combine:

generative AI for early assessment and summaries;
CAL for prioritising formal responsiveness review;
targeted human review for privilege and difficult legal decisions; and
documented quality control before production.

The appropriate approach will depend on the size of the matter, the review objective and the level of defensibility required.

Confidentiality May Be the Main Constraint

The Federal Court’s guidance places particular emphasis on confidential, privileged, suppressed and private information.

Uploading discovery material into a generally accessible AI tool may expose it to storage, processing or reuse outside the legal team’s control. This could conflict with confidentiality orders, privilege obligations or restrictions on documents obtained through compulsory court processes. A closed or enterprise AI environment may reduce some risks, but it does not automatically make every use acceptable.

Legal teams should understand:

where the information is processed and stored;
whether prompts or documents are retained;
whether customer data is used to train models;
who can access the information;
how long the data remains available; and
whether the system provides an audit record.

These questions should be addressed before documents are uploaded.

When Must AI Use Be Disclosed?

The Practice Note does not require disclosure for every AI-assisted search, summary or review exercise.

Disclosure is specifically required where generative AI has been used:

to summarise or analyse information on which a witness relies;
to create multimedia presented to the Court; or
in another way reasonably expected to affect the admissibility or use of evidence.

The Court may also order disclosure in other circumstances.

As a practical governance measure, legal teams should consider recording:

the AI tool and purpose;
the documents or data provided;
the prompts or review instructions;
the human checking undertaken;
any errors identified; and
the final decisions made by reviewers.

This record may help the team explain and defend its process if the use of AI is later questioned.

What Legal Teams Should Do

Define the review objective

Decide whether AI is being used for early assessment, factual extraction, responsiveness review, privilege identification or production decisions.

Use an appropriate platform

Avoid uploading confidential discovery material into public AI services. Confirm the security, contractual and data-handling controls of any approved system.

Preserve the original evidence

AI-generated summaries and classifications should remain linked to the original documents and metadata.

Retain human responsibility

Experienced reviewers should confirm important responsiveness, privilege and production decisions.

Validate and document the process

Test the workflow against known documents and selected samples. Record the tools, instructions, limitations and quality-control steps used.

Why This Matters in New Zealand

The Federal Court of Australia’s Practice Note does not apply directly to New Zealand proceedings. However, it provides a useful regional indication of how courts may approach AI-assisted litigation. It accepts that AI can improve efficiency while making clear that legal teams remain accountable for accuracy, confidentiality and compliance.

New Zealand organisations using generative AI in investigations or document review should expect similar questions from courts, regulators, clients and opposing parties.

The issue is not simply whether AI was used. The more important question is whether it was used through a controlled, explainable and defensible process.

The Takeaway

Generative AI can make document review faster and provide valuable early insight into large collections.

It should not be treated as a replacement for sound discovery methodology, established Technology-Assisted Review or experienced legal judgement.

The strongest approach combines secure technology, clearly defined review objectives, human oversight, validation and proper documentation.

How Forensic Tech Can Help

Forensic Tech assists legal teams and organisations with defensible document review and electronic discovery.

This includes evidence collection, data processing, Technology-Assisted Review, Continuous Active Learning, AI-assisted analysis, privilege workflows, review quality control and secure, documented productions for litigation, investigations and regulatory matters.