The Solicitors Regulation Authority has issued an important warning about AI, confidentiality and legal privilege. Here is what it means for ordinary people using AI to challenge powerful institutions.
Steve Conley | Get SAFE | 30 October 2026
AI is changing the balance of power
For years, people challenging banks, insurers, pension providers and other powerful institutions have faced a familiar problem.
The institution has lawyers, compliance departments, specialist knowledge and extensive resources.
The individual often has a folder of documents, an unfamiliar complaints process and very little idea where to begin.
AI is beginning to change that.
A person can now use AI to organise correspondence, construct a chronology, identify contradictions, prepare questions and understand complicated explanations.
Someone who struggles to write a complaint can produce a coherent account of what happened.
Someone who feels overwhelmed by hundreds of documents can begin making sense of them.
Someone who previously depended entirely on a professional can become an informed participant in their own case.
That is an important restoration of human agency.
But using AI responsibly requires understanding one critical distinction.
Not every document you possess is necessarily yours to share.
What the SRA has warned about
On 17 August 2026, the Solicitors Regulation Authority published a warning notice about the misuse of AI.
It highlighted two particular concerns:
- AI generating inaccurate information, including fictitious legal cases and references.
- Confidential client information being entered into AI systems without adequate safeguards.
The second concern is especially important.
In UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), the Upper Tribunal raised serious questions about entering confidential legal documents into publicly accessible AI systems.
The SRA warned that inappropriate use could compromise confidentiality and potentially result in the permanent loss of legal professional privilege.
This guidance is directed at solicitors and regulated legal practices.
But the underlying risks are relevant to everyone using AI in connection with complaints, disputes or litigation.
We should understand those risks without allowing them to become another reason to discourage citizens from using technology.
Three different kinds of documents
It helps to distinguish three categories.
1. Ordinary evidence
Examples include:
- Your bank statements.
- Letters addressed to you by a financial institution.
- Your own account of what happened.
- Publicly available regulatory publications.
- Published court judgments.
- Correspondence you are entitled to use and disclose.
These documents can often be useful in AI-assisted case preparation.
However, personal information still needs protecting. A document being in your possession does not automatically mean every use or disclosure is permitted.
Before uploading, check the AI provider’s privacy arrangements, data retention policies and treatment of uploaded material.
2. Confidential information
Some documents contain information that should not be shared freely.
Examples might include:
- Another person’s medical or financial records.
- Commercially sensitive agreements.
- Confidential settlement negotiations.
- Internal company correspondence.
- Documents containing third-party personal information.
The important question is whether you have permission or another lawful basis to disclose the information to the particular AI service.
A paid AI subscription does not, by itself, make that disclosure safe.
Nor does a promise that the AI will keep everything confidential.
You need to understand what happens to the information after it leaves your computer.
3. Legally privileged or disclosure-restricted documents
These require particular care.
Examples include:
- Confidential legal advice from your solicitor.
- Communications protected by legal professional privilege.
- Documents subject to a court confidentiality order.
- Material disclosed during litigation under restrictions on its use.
- Information covered by a contractual non-disclosure obligation.
These restrictions are not all the same.
Legal professional privilege protects certain confidential lawyer-client communications. Court orders and contractual confidentiality obligations operate under different legal rules.
A document may be subject to more than one restriction.
The safest starting point is simple: do not upload legally privileged or disclosure-restricted material into an external AI service until you have established that doing so is permitted.
If you have a solicitor, ask them first.
A secure AI system does not override a court order
This is an important distinction.
An AI provider may promise encryption, restricted access, short retention periods and protection against model training.
Those are valuable safeguards.
But they do not automatically give you permission to disclose information you are otherwise prohibited from sharing.
Similarly, a private AI workspace does not automatically create or preserve legal professional privilege.
What matters is the nature of the information, the restrictions governing it, the recipient and the safeguards surrounding the disclosure.
In some circumstances, lawyers can arrange appropriately controlled third-party processing of privileged material. But that requires careful consideration.
Technology cannot simply declare itself exempt from legal restrictions.
Can you use AI without uploading the restricted document?
Often, yes.
Suppose you have received confidential legal advice about a dispute.
You might be tempted to upload the advice and ask:
“Is my solicitor right?”
That could create unnecessary risks.
A safer approach may be to ask general questions without revealing the privileged advice or its substance.
For example:
“What questions should someone ask a solicitor when evaluating the enforceability of a commercial loan agreement?”
Or:
“What factors might be relevant when a business finances the purchase of contractual servicing rights that terminate when an agency relationship ends?”
These questions can improve your understanding without necessarily disclosing confidential legal communications.
However, even a summary can reveal protected information. Rewriting privileged advice in your own words does not automatically remove privilege or disclosure restrictions.
When uncertain, seek clarification before sharing.
AI can invent legal authorities
The SRA’s second major warning concerns accuracy.
AI can generate convincing legal explanations supported by cases that do not exist.
It may also cite a real judgment but misrepresent what the judge decided.
That is dangerous when someone is preparing a complaint, witness statement or court submission.
A useful discipline is to separate five things:
- What the documents establish.
- What one party alleges.
- What the evidence reasonably suggests.
- What remains uncertain or disputed.
- What requires legal interpretation.
AI can help organise all five.
But it should not silently turn an allegation into a fact or a possibility into a legal conclusion.
If AI refers to legislation, regulations or judgments, verify those references against authoritative sources.
Useful starting points include:
If a document is intended for court proceedings, the person submitting it remains responsible for its accuracy. Legal representation or other appropriate review may be necessary.
AI can help you prepare your case. It cannot take responsibility for your case.
What this means for Get SAFE and Goliathon
At Get SAFE, we believe access to justice should not depend on someone’s ability to afford unlimited professional assistance.
Financial exploitation can leave people distressed, confused and afraid of challenging the institutions involved.
Discouraging them from using AI would remove an increasingly powerful means of understanding what happened and asking better questions.
But encouraging AI use carries responsibilities of its own.
That is why we are developing Goliathon around three principles:
Private by design. Evidence by reference. Decisions remain human.
Our development priorities include stronger controls over access to case information, clearer warnings about restricted documents, improved evidence traceability and more explicit distinctions between established facts and unverified assertions.
These are development objectives, not claims that Goliathon is already approved for processing legally privileged material.
Until the relevant safeguards are verified, people should not upload privileged or disclosure-restricted documents to Goliathon.
The objective is to help citizens organise their own evidence and engage more effectively with professional advisers, ombudsmen and courts.
It is not to replace lawyers or make unsupported promises about legal protection.
Five questions before you upload a document
Before putting sensitive evidence into any AI system, ask yourself:
- Am I entitled to share this document? Possession does not necessarily establish permission.
- Does it contain confidential information about someone else? Consider whether that information needs to be included.
- Could it be legally privileged or disclosure-restricted? If so, pause and seek appropriate advice.
- Where will the information go? Check how the provider processes, stores, retains and deletes it.
- Will I verify the AI’s conclusions? Treat the output as assistance with analysis, not proof.
These questions will not resolve every legal issue, but they can prevent avoidable mistakes.
The real lesson: capability and caution belong together
There is a danger that warnings about AI will be interpreted as warnings against citizens using it.
That would be the wrong lesson.
The financial services establishment already uses sophisticated technology, specialist knowledge and professional support to protect its interests.
Citizens should be able to develop comparable capabilities to understand and challenge institutional decisions.
The SRA’s warning gives us a reason to build safer systems and use them more intelligently.
It does not justify making ordinary people dependent on professional gatekeepers for every question they want to ask.
There is a difference between protecting someone from risk and preventing them from exercising agency.
We should understand that difference.
Use AI to strengthen your understanding. Protect information you are not entitled to disclose. Verify important claims. Keep control of your decisions.
Because access to justice should mean more than access to a lawyer.
It should include the capability to understand your own case.
About Get SAFE
Get SAFE supports people affected by financial exploitation, helping them regain clarity, confidence and control.
Our approach combines practical education, independent support and AI-assisted tools that help people organise evidence, understand their options and participate more effectively in decisions affecting their lives.
Self-directed by default. Supported when it matters.
This article provides general information, not legal advice. The application of legal professional privilege, confidentiality obligations and court restrictions depends on the particular circumstances. Where restrictions may apply, obtain advice from an appropriately qualified legal professional before disclosing material.
