The AI client: navigating privacy and privilege in family proceedings

The landscape of family law is shifting. While the profession debates the structural integration of AI into our firms, our clients have already integrated AI into their cases and daily lives. Armed with large language models and autonomous AI agents, clients are increasingly bypassing or supplementing traditional legal correspondence and advice. Many of us will have seen evidence of clients using AI to analyse the other parties’ evidence, to draft emails and to scrutinise the advice we give to them.

For those in the field of family law, this presents a distinct set of challenges. While AI can undoubtedly lower costs and empower individuals, particularly those who are neurodiverse, its unchecked use poses risks to confidentiality, legal professional privilege and the integrity of the court process.

The privacy problem

Family law disputes often deal with the most sensitive categories of personal information: eg financial records, health data and private details concerning children.

When a client uploads a document (such as a witness statement or a Form E) into an open-source or public AI tool like ChatGPT or Claude, they may not be aware of where that data goes. In most public models, data inputs are processed on external servers and utilised to train future iterations of the software. Some models allow the user to turn this function off, others promise time-limited use, but many users will be unaware of the impact of their use or of their ability (if any) to control it.

In the case Sapphira discusses in her article, UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), the Upper Tribunal warned of the impact of this behaviour. Although an immigration matter, the court’s comments on AI and data security apply equally to family law. The Tribunal highlighted that uploading confidential documents into an open-source AI tool places that information directly into the public domain. Consequently, the act constitutes a breach of client confidentiality and a waiver of legal professional privilege. It may even constitute contempt of court.

Unrealistic expectations and the hallucination hazard

An additional risk that clients and the family justice system face is the apparent certainty with which generative AI answers a user’s query. The response to a query will read as fact with no room for doubt. Consequently, users are sometimes driven to wholly unrealistic positions and expectations as to outcome. Generative AI works on probability, not accuracy. It is highly adept at creating linguistically persuasive arguments that can be unrealistic or even entirely untethered from reality.

In Re A, B, C, D (Extension of assessment; Use of AI: Hallucinations) [2026] EWFC 71 (B) the Family Court considered a scenario where a therapist acting as a lay advocate for a friend in Children Act proceedings presented a skeleton argument containing four entirely fabricated case citations generated by AI. The Recorder took the significant step of publicly naming the individual, albeit accepting that they had no intention of misleading the court.

The court in Re A (understandably) expressed sympathy for the fact that litigants in person naturally turn to AI for assistance, but emphasised that all parties owe a strict duty to ensure the material they present is accurate.

What should family practitioners do?

As family solicitors, we cannot police our clients’ digital habits, but we can manage our risk and protect their positions.

• Update Client Care letters: ensure your terms of engagement explicitly warn clients about the dangers of open-source AI. Explicitly state that uploading any case material or correspondence to a public AI tool can destroy legal privilege and breach court confidentiality rules.
• Audit “client-drafted” materials: if a client provides a remarkably polished, legally structured narrative or a heavily managed emotional email, question its origin. Examine the text for unusual phrasing or non-jurisdictional legal concepts (such as American terminology like “custody” or “visitation”).
• Differentiate closed versus open tools: educate clients that if they must use digital organisational tools, they should ensure the platform guarantees absolute data privacy and explicitly states that inputs are not used for model training.

AI is here to stay, and used responsibly within closed, secure legal frameworks, it represents an opportunity to better the service we provide to our clients. However, when relied on by clients navigating the alien world of family law, it risks turning a private dispute into a public data breach.

[email protected]