A recent judgment published by the Upper Tribunal Immigration and Asylum Chamber has highlighted concerns about the use of open-source artificial intelligence tools in legal practice, warning lawyers against putting client documents into ChatGPT. The ruling, arising from two immigration matters involving fictitious case citations, contains explicit warnings about the use of tools such as ChatGPT with client documents, and confirms that doing so may constitute a breach of client confidentiality, a waiver of legal privilege, and a reportable data breach.
This guidance note sets out what the judgment means for your firm, why existing AI use policy wording may not go far enough, and the immediate steps you should take.
What the Tribunal Said
Judge Fiona Lindsley, sitting in the Upper Tribunal Immigration and Asylum Chamber, was unambiguous:
“Uploading confidential documents into an open-source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege, and any such conduct might itself warrant referral to the SRA and should, in any event, be referred to the Information Commissioner’s Office.”
In the case before the Tribunal, a solicitor had uploaded client emails and Home Office decision letters into ChatGPT, not to generate legal submissions, but simply to summarise content and improve draft client communications. The Tribunal confirmed this was still a data breach, irrespective of the intended use.
The solicitor was required to notify affected clients, self-report to the Solicitors Regulation Authority (SRA), and the Immigration Advice Authority (IAA). The Tribunal confirmed it would have referred him had he not already done so.
Another key issue was in regard to “hallucinated case references” being put before the tribunal, in which citations of cases which did not exist and placed false information before the court.
“A solicitor or other legal professional who delegates their work to another fee-earner remains responsible for the supervision of their work and for ensuring its accuracy. Such supervisors must ensure that fee-earners under their supervision are aware of the dangers of using non-specialist Artificial Intelligence (AI) for legal research and drafting. Failures to do so, or to undertake appropriate checks on the drafting of fee-earners is likely to result in a referral to the Solicitors Regulation Authority (SRA) or other professional body. A supervisor who fails to ensure that the work of a more junior fee-earner does not contain false cases or citations is likely to be more culpable than a lawyer who fails to ensure that his own work is free from such “hallucinations”.
Information on this judgement was published by Legal Futures, Source: Solicitor faces probe after putting client documents into ChatGPT.
Why This Matters for Your Firm’s Current AI Policy
Following discussions with our compliance managers across client firms, we have identified some common areas of potential vulnerability in how AI use is currently governed and permitted. A number of firms permit the use of ChatGPT, and similar consumer-grade AI tools, within their AI policies, but with qualifying language that does not adequately address the data risk.
For example, some policies qualify usage by:
“Initial analysis of documents only”
Ensure your policies are clear about which documents are allowed. Without qualifying language, documents, including a client letter, a Home Office decision, or a case file, uploaded to ChatGPT enters a public platform. The purpose of the upload is irrelevant to the breach. Client confidentiality is compromised at the point of upload, regardless of what the tool is being asked to do with that document.
“Population of documents / draft documents”
Uploading documents for the purpose of drafting or populating precedents carries exactly the same risk. If the document contains client-identifiable or legally privileged information and it is processed by an open-source model, confidentiality is breached.
The critical point to remember is that it is not the task being performed that creates the breach; it is the act of uploading client data to a platform that processes it externally. No qualifying language, however well-intentioned, makes that safe.
Open-Source vs Closed-Source: Understanding the Distinction
The Tribunal itself drew a practical distinction between the two categories of AI tool, expressly noting that closed-source models, specifically citing Microsoft Copilot as an example, can be used for tasks such as summarising documents without the same risks.
Open-source/consumer AI tools (such as ChatGPT, Google Gemini in its consumer form, and Meta AI) process data externally, may use inputs to train future versions of the model, and effectively place information in the public domain. These tools must not be used with any client data.
Closed-source/enterprise AI tools operate within a controlled environment, either the firm’s own infrastructure or a vendor’s contracted environment, with data processing agreements in place that prevent client data from being shared or used for model training. These tools can be used appropriately, provided the relevant contractual protections are confirmed.
Our strong recommendation, consistent with the Tribunal’s guidance and the consensus of our compliance managers, is that firms should err firmly on the side of caution: if a tool is not confirmed as closed-source with appropriate data protection controls, it should not be used with client data.
Immediate Actions for Your Firm
1. Review and update your AI use policy
If your current policy permits the use of ChatGPT or other consumer AI tools in any context involving client data, even with qualifying terms, it should be updated. The policy should clearly distinguish between approved closed-source tools with data processing agreements in place, and prohibited open-source tools for any task involving client information. Permitted use of open-source tools should be restricted to internal tasks that involve no client data whatsoever.
2. If ChatGPT accounts exist: disable the learning function immediately
If staff are currently using ChatGPT and an immediate prohibition cannot be enforced straight away, the most important interim step is to disable the model’s ability to use your data for training purposes. This does not eliminate the underlying data risk, data is still being processed externally, but it reduces the risk of client information being incorporated into future model training.
To do this: go to your ChatGPT account, select Settings, then Data Controls, and toggle off “Improve the model for everyone.” This must be done on each individual account. If multiple members of staff have their own ChatGPT accounts, each one must be updated separately. Further information can be accessed here: https://help.openai.com/en/articles/7730893-data-controls-faq
This is a safeguard, but not the most robust solution, as it does not guarantee that ChatGPT is a safe platform for loading client data.
3. Issue an all-staff briefing
The Tribunal was explicit that a supervisor who fails to ensure their staff are aware of AI risks is likely to be considered more culpable than the member of staff who makes the error. All fee-earners and support staff should be briefed on the distinction between open and closed-source AI tools, and should receive written confirmation that uploading any client document, email, letter, or file to consumer AI tools is prohibited.
Firms should also be aware that AI tools are accessible via any standard Google search. It is not sufficient to assume that staff are not using them simply because no AI software has been formally installed on firm systems.
4. Consider your reporting obligations
If you are aware, or become aware, that client documents have previously been uploaded to ChatGPT or similar tools, consider your reporting obligations under your data protection policies, consider whether a report to your regulator is required, and consult the ICO. The Tribunal specifically directed that this should occur where such conduct has taken place.
AI Tools with Closed Source
We know of a number of closed-source AI tools that are already in use across our firms, with some of the most common being:
Claude AI (Anthropic) — Enterprise licence Where firms are using Claude via an enterprise licence or API integration, data is not used for model training and remains within the contracted environment. Firms should confirm the specific plan in use and review their data processing agreement.
LexisNexis AI LexisNexis operates as a closed environment with established data processing agreements. Its AI tools are built for legal professionals and do not expose client data to public models. Verify your contract terms are current.
LEAP (Practice Management AI) LEAP’s AI functionality operates within the platform’s existing data environment. Client data processed through LEAP’s AI tools remains within your firm’s system, subject to LEAP’s data processing terms.
In all cases, firms should have a current data processing agreement in place with any AI tool provider. If you are unsure whether one exists for a tool your firm is currently using, contact your supplier directly or speak to your RACO compliance manager.
Summary Checklist
- Review your AI use policy and remove or revise any permission for open-source AI tools to be used with client data
- If ChatGPT accounts exist, disable the model improvement/learning setting on each account immediately
- Issue an all-staff briefing confirming the prohibition on uploading client data to consumer AI tools
- Confirm data processing agreements are in place for all AI tools currently in use, including Claude, LexisNexis, and LEAP
- If past uploads to open-source tools are known to have occurred, consider reporting obligations to clients, the SRA, and the ICO
- Ensure supervisors understand their personal responsibility for the AI practices of those they oversee
Further Compliance Support
If your firm requires assistance reviewing, drafting, or updating your AI use policy, assessing your current tool landscape, or preparing staff communications and training, please contact your RACO compliance manager.
To ask any questions about the Regulation and Compliance Office and how we can support your Law Firm, please call us on 0203 985 8550, complete the online enquiry form below, or email us info@regulationandcomplianceoffice.co.uk
