What do we mean by open AI and closed AI?

The language can be confusing. People often use “open AI” to mean several different things:

• AI models where the code is open;
• AI models where the model weights are available;
• AI systems that can be run locally rather than through a private cloud service;
• AI tools where the training data and methodology are transparent;
• AI tools that can be inspected, tested, adapted or independently audited; or
• AI that is simply marketed as “open”, even where important parts remain hidden.

This matters because open-source AI, open-weight AI and transparent AI are not always the same thing.

For family law legal tech, the question is not ideological. It is practical:

Which model gives the safest, fairest, most accountable support for people dealing with family court?

Why family law is different

Family law is not like ordinary commercial work. It involves children, safeguarding, domestic abuse, emotional distress, personal histories, disputed allegations and confidential court material.

A family law AI tool may be asked to assist with:

• chronologies;
• position statements;
• witness statement structure;
• Cafcass report review;
• domestic abuse allegation schedules;
• child arrangements issues;
• bundle organisation;
• fact-finding preparation;
• summaries of messages, emails and disclosure;
• identifying missing evidence;
• drafting questions for hearings;
• explaining court orders in plain English; and
• helping litigants in person understand procedure.

These are high-risk tasks. A mistake may not merely inconvenience someone. It may affect how risk is presented, how a child’s welfare is understood, whether domestic abuse is properly identified, or whether a litigant in person feels falsely reassured.

The case for closed AI in family law legal tech

Closed AI systems are often criticised because users cannot see fully inside them. But they may have important advantages, particularly where the provider has invested heavily in safety, security, reliability, infrastructure, monitoring and user support.

In family law legal tech, closed AI may offer:

• stronger infrastructure — enterprise-grade hosting, uptime, resilience and support;
• better usability — interfaces that ordinary users can actually understand;
• advanced model capability — strong drafting, summarisation and reasoning support;
• centralised safety controls — provider-level guardrails, abuse monitoring and updates;
• contractual protections — enterprise agreements, data processing terms and service-level commitments;
• rapid updates — improvements can be deployed quickly by the provider;
• support for integrations — document systems, CRMs, practice management platforms and secure legal workflows; and
• lower technical burden — law firms and support organisations do not need to host or maintain their own models.

For many small law firms, charities, McKenzie Friend services, legal support providers and litigants in person, a closed AI tool may be more realistic than building and maintaining a local AI system.

But the trade-off is trust. If the model is closed, users may not know exactly how it was trained, what data influenced it, how it handles bias, or why it produced a particular answer.

The case for open AI in family law legal tech

Open AI appeals to many legal technologists because it promises greater transparency, independence and control.

In family law, open or locally deployable AI could offer:

• greater auditability — researchers and developers may be able to test behaviour more closely;
• local deployment — sensitive material may be processed within a controlled environment rather than sent to an external cloud service;
• customisation — models can potentially be adapted for family law procedure, domestic abuse terminology and litigant in person support;
• reduced vendor lock-in — organisations are not entirely dependent on one commercial provider;
• cost control — open models may reduce long-term cost for public-interest projects;
• public-interest innovation — universities, charities, legal clinics and access-to-justice groups can build tools without waiting for commercial providers;
• independent testing — bias, hallucination and safeguarding risks can be examined more openly; and
• sovereignty and control — courts, public bodies or legal charities may prefer systems they can govern directly.

Open AI may be particularly important for public-interest legal technology. If access-to-justice tools are controlled entirely by private providers, there is a risk that family justice innovation becomes dependent on commercial priorities rather than public need.

The risks on both sides

Neither open AI nor closed AI is automatically safe. Both can be used well. Both can be used badly.

The right question is not “which is good and which is bad?” The right question is:

What safeguards are in place for this specific family law use case?

What this means for litigants in person

Litigants in person are already using AI. Some use it to explain orders, draft emails, summarise evidence, prepare statements or understand court language.

That can be helpful. It can also be risky.

A litigant in person may not know:

• whether the tool stores their information;
• whether uploaded documents may be used to improve a model;
• whether the answer is accurate under family procedure;
• whether the tool is inventing law or cases;
• whether confidential family court material can be entered safely;
• whether the tool understands domestic abuse dynamics;
• whether the output is too emotional, too aggressive or procedurally inappropriate; or
• whether they should seek urgent legal advice instead.

For litigants in person, AI should be a support tool, not a decision-maker. It can help create order from chaos, but it cannot understand your child, your risk, your judge or your evidence in the way a properly informed human professional can.

What this means for solicitors, barristers and McKenzie Friend support

Legal professionals and litigation support providers need to think carefully about what kind of AI they use and for what purpose.

For professional users, the key questions include:

• Is client consent required before using AI on their material?
• Is the data being uploaded to a third-party system?
• Is the tool covered by a proper data processing agreement?
• Can confidential, privileged or sensitive material be used safely?
• Can outputs be checked by a competent human?
• Is the model being used for administrative support or legal reasoning?
• Is the tool suitable for domestic abuse and safeguarding material?
• Is there an audit trail?
• Who is responsible if the output is wrong?
• Can the organisation explain its AI use to clients and the court?

In practice, the safest immediate uses of AI in family law are likely to be:

• document organisation;
• drafting neutral chronologies;
• identifying missing documents;
• creating first-draft hearing preparation notes;
• turning emotional narratives into structured issue lists;
• summarising long message threads, subject to confidentiality controls;
• plain-English explanation of procedural terminology; and
• internal workflow support.

The higher-risk uses are:

• predicting case outcomes;
• assessing witness credibility;
• ranking parental risk without expert oversight;
• generating legal advice without review;
• drafting allegations without evidential checking;
• summarising children’s wishes and feelings without context;
• analysing domestic abuse dynamics without specialist knowledge; and
• producing court-ready documents with no human review.

What this means for the family courts

The courts will increasingly encounter AI-generated material. Litigants in person may file AI-assisted statements. Lawyers may use AI to summarise bundles. Judges may use AI in limited administrative or research-support contexts, subject to judicial guidance.

The family court will therefore need a practical approach, not panic and not blind enthusiasm.

The court may need to ask:

• Was AI used to prepare this document?
• Has the party checked the content personally?
• Are there invented cases, inaccurate rules or unsupported allegations?
• Has confidential material been handled appropriately?
• Is the document still the party’s own evidence?
• Has AI made the material clearer, or has it distorted the party’s voice?
• Is the use of AI creating unfairness between represented and unrepresented parties?

This is where family law legal tech must be designed carefully. The goal should not be to make every litigant sound like a barrister. The goal should be to help people present relevant facts, evidence, chronology and safeguarding concerns in a way the court can understand.

The future: hybrid, accountable and human-led

The future of family law legal tech is unlikely to be purely open or purely closed.

The better model is likely to be hybrid:

• closed, secure systems for some high-capability tasks;
• open or locally hosted models for sensitive document analysis;
• specialist family law workflows designed by people who understand court process;
• clear user warnings and limits;
• human review before anything is filed or relied upon;
• audit trails for professional users;
• privacy-by-design architecture;
• plain-English support for litigants in person;
• specialist safeguarding prompts; and
• transparent governance about what the tool can and cannot do.

Family law AI should be judged by practical outcomes:

• Does it help people understand the process?
• Does it protect confidentiality?
• Does it reduce overwhelm?
• Does it improve the quality of evidence organisation?
• Does it avoid making unsafe assumptions?
• Does it support, rather than replace, human judgment?
• Does it make access to justice better for people who cannot afford representation?