Financial advice firms using ChatGPT, Microsoft Copilot, Claude or other general-purpose AI tools need to understand what happens to client information once it enters those systems. The main privacy risks involve unauthorised data sharing, unclear processing locations, retention arrangements, third-party access and a lack of consistent organisational controls. Enterprise licences can reduce certain risks, but they do not automatically make every AI workflow compliant with UK GDPR or FCA expectations. Before allowing AI into your advice process, you need to understand how information is handled and who remains responsible for protecting it.
For advisers and paraplanners, this is particularly important because financial planning documents often contain substantial amounts of personal and confidential information.
A typical fact find may include employment details, income, savings, pension arrangements, financial commitments, family circumstances and retirement objectives. Meeting notes can also contain sensitive information about a client’s health or personal circumstances.
When this information is entered into an AI tool, your firm needs to understand how it is processed, even if the employee is simply asking for help preparing a suitability report.
The question is not whether AI should be used in financial advice. It is whether your firm has sufficient visibility and control over the information being shared.
1. What Client Information Is Being Uploaded, and Where Is It Processed?
Before allowing employees to use AI with client information, you need to establish exactly what data enters each platform and where it is processed. This includes information pasted into prompts, uploaded documents, meeting transcripts and any data accessed through connected applications. You should also understand which organisations process that information and whether international data transfers are involved.
The challenge is that AI use can seem relatively harmless during everyday work.
Imagine a paraplanner preparing a suitability report for a client approaching retirement. They have a fact find, pension statements and notes from a recent meeting.
Rather than manually reviewing every document, they upload the files to an AI assistant and ask it to summarise the client’s circumstances.
Within seconds, they have a useful overview.
However, those documents may contain considerably more personal information than was necessary to produce the summary.
The pension statement might include an account reference, the fact find may contain the client’s home address, and the meeting notes could refer to a recent medical diagnosis.
The paraplanner’s intention was simply to save time, but the firm has now introduced another organisation into the processing of potentially sensitive information.
What information should your firm be particularly careful with?
Not all client information carries the same level of risk, but financial advice firms routinely process data that deserves careful handling.
This can include:
- Personal Identification: Client Names, Addresses, Dates Of Birth And Contact Details.
- Financial Information: Income, Savings, Investments, Pension Values And Outstanding Debts.
- Account Information: Policy Numbers, Account References And Financial Provider Details.
- Family Circumstances: Dependants, Beneficiaries, Relationships And Estate Planning Information.
- Health Information: Medical Conditions, Disabilities And Health-Related Circumstances Relevant To Financial Planning.
- Client Vulnerability Information: Details Of Circumstances That May Affect A Client’s Ability To Understand Or Act On Financial Advice.
Health information is particularly important because it can qualify as special category data under UK GDPR, requiring additional legal conditions for processing.
Financial information is not automatically special category data, but it is still personal data when it relates to an identifiable individual.
Does removing a client’s name make the information anonymous?
Not necessarily.
Removing a name or account number can reduce privacy risks, but it does not automatically anonymise a document.
Consider a client description containing their age, occupation, location, pension value and a particular family circumstance.
Even without their name, the combination of details might make the individual identifiable, particularly when combined with other information.
This distinction matters because pseudonymised information remains personal data where identification is still possible using additional information.
If you are using general-purpose AI for internal work, consider whether the task can be completed using genuinely anonymous examples or information that does not identify a client.
Where identifiable information is necessary, you need an approved processing arrangement and appropriate safeguards.
Where is AI processing taking place?
A tool’s UK availability, UK office or UK-based data centre does not automatically mean that all processing remains within the UK.
You may need to consider the location of the contracting organisation, subprocessors, support access, backups and other processing activities.
International transfer obligations depend on the specific arrangements, not simply the physical location of a server.
The Information Commissioner’s Office (ICO) explains these distinctions in its guidance on international transfers of personal information.
Before approving a provider, ask for clear documentation about how information moves through its systems.
2. Who Can Access Client Information, and How Long Is It Retained?
You need to know who can access information submitted to an AI platform, why access might be required and how long the information remains available. This includes employees, administrators, technology providers and authorised subprocessors. Retention and deletion arrangements should cover prompts, uploaded files, generated outputs, backups and relevant technical logs, rather than focusing only on the visible chat history.
One of the easiest mistakes to make is assuming that deleting a conversation removes every copy of the information involved.
Depending on the platform, removing a chat from the user interface may not immediately remove information from backups, security logs or other systems.
Retention policies can also vary between personal accounts, business subscriptions, enterprise products and API services.
You should therefore examine the specific terms applying to the product and configuration your firm intends to use.
Questions to ask about access and retention
| Privacy consideration | Question your firm should ask | Why it matters |
| Employee access | Who can view prompts, documents and generated reports? | Helps prevent unnecessary internal access to confidential information |
| Provider access | Can authorised provider personnel access submitted information, and under what circumstances? | Clarifies third-party access and oversight arrangements |
| Model training | Is client information used to train or improve AI models? | Establishes whether information may be reused beyond the immediate task |
| Data retention | How long are prompts, files, outputs and logs retained? | Supports data minimisation, storage limitation and retention policies |
| Deletion | What happens when a user deletes information or the contract ends? | Helps establish whether and how information can be removed |
| Subprocessors | Which additional organisations handle the information? | Identifies further processing relationships and potential transfer risks |
These questions should be answered through appropriate contractual and technical documentation, not assumptions based on marketing descriptions.
Why ‘not used for training’ is only part of the answer
Many organisations understandably prioritise whether their client data will be used to train an AI model.
That is an important consideration, but it addresses only 1 aspect of data protection.
A platform may exclude customer data from model training while still processing or retaining it to provide the service, maintain security, investigate misuse or meet legal obligations.
Those activities may be legitimate, but your firm still needs to understand them.
For instance, if a provider states that prompts are not used for training, ask whether the information is retained for another purpose and whether your firm can control that retention.
You should also establish whether the answer applies to every employee account and feature being used.
An organisational subscription does not necessarily govern an employee’s separate personal AI account.
3. Why an Enterprise Licence Solves Part of the Problem, but Not All of It
Enterprise AI licences can provide useful protections, including centralised administration, identity management, access controls and contractual commitments concerning data handling. However, they do not automatically ensure that employees upload appropriate information, follow approved procedures or produce accurate client documents. Your firm still needs a clear policy, suitable technical settings, staff training and defined responsibilities for checking AI-generated work.
The difference between consumer AI services and enterprise products can be significant.
Business-focused subscriptions may offer greater control over how organisational information is processed, who can use the system and which applications employees can connect.
Those features can make a platform more appropriate for workplace use.
However, an enterprise licence does not remove the need to assess individual workflows.
Consider a firm that purchases enterprise AI access for its paraplanning team.
The provider may offer appropriate contractual safeguards and administrative controls, but an employee might still upload an unnecessarily detailed fact find when a short, anonymised extract would have been sufficient.
Another employee might connect an external application that has not been approved by the firm.
A colleague could use the platform to draft a suitability report without checking that the figures and explanations match the client file.
None of these problems is necessarily solved by the licence itself.
Security and governance address different risks
Security controls focus on protecting information from unauthorised access, disclosure or other security threats.
Governance considers how the technology is used and whether the organisation can demonstrate that its processes are appropriate.
For financial advice firms, both are necessary.
A secure AI assistant may be appropriate for drafting general internal communications. That does not automatically make it suitable for processing a client’s full financial circumstances or preparing regulated advice documentation.
The better approach is to assess AI tools against the specific activities your firm wants to undertake.
Why purpose-built reporting software can offer a more controlled approach
General-purpose AI tools are designed to support many different tasks across different industries.
Financial advice reporting has more specific requirements.
Your firm needs to maintain approved wording, use appropriate source information and produce documents that reflect its reporting standards.
Purpose-built suitability report software can help bring those requirements into a more structured workflow.
Ammonite’s Planbot, for example, is designed around the needs of UK financial advisers and paraplanners.
Rather than relying on employees to repeatedly create their own prompts, firms can use existing report templates and control the wording, sections, tone and formatting of their documents.
Ammonite states that Planbot does not store the sensitive client data used to populate reports and does not use client information to train AI models.
Its published information also identifies UK-based Google Cloud hosting and Cyber Essentials certification.
These are relevant considerations when evaluating a reporting platform, although your firm should still review the provider’s current contractual, technical and security documentation before introducing any system into its workflow.
The aim should be to select technology that supports your existing responsibilities rather than introducing another process that employees must manage independently.
4. Meeting Your UK GDPR and FCA Obligations When AI Enters the Workflow
UK GDPR applies when your firm processes personal data using AI, and existing FCA requirements continue to apply to regulated activities. You must have an appropriate lawful basis, understand the purpose of processing, protect information and maintain suitable governance arrangements. AI does not create an exemption from these responsibilities, nor does it automatically require client consent for every use.
This is particularly relevant because AI systems may process information in ways that are less obvious than traditional document software.
A financial advice firm will generally need to consider its role as a data controller and the responsibilities of any technology provider acting as a processor.
The exact allocation of responsibilities depends on the nature of the service and how the parties use the information.
UK GDPR and the use of client information
UK GDPR requires organisations to follow principles covering lawfulness, fairness and transparency, purpose limitation, data minimisation, accuracy, storage limitation, security and accountability.
These principles apply to AI-assisted document preparation just as they apply to other processing activities.
For example, suppose your firm collects a client’s financial information to provide regulated advice.
Using an AI tool to assist with the preparation of advice documentation may be compatible with that purpose, depending on the circumstances.
However, allowing the information to be reused for an unrelated purpose, such as training a provider’s general model, raises separate considerations.
Your firm should understand whether the processing is lawful, necessary and consistent with the information provided to clients.
The ICO’s guidance on accountability and governance in AI is a useful reference when assessing these responsibilities.
Does your firm need a data protection impact assessment?
A data protection impact assessment (DPIA) is required where the proposed processing is likely to result in a high risk to individuals’ rights and freedoms.
AI systems can introduce risks relating to the scale, sensitivity and complexity of personal data processing, but the assessment needs to reflect the actual use case.
Your firm should therefore evaluate the nature of the information, how the system operates, what could go wrong and whether appropriate safeguards are available.
For instance, using AI to help improve a generic training document presents different risks from processing detailed client financial and health information.
Even where a DPIA is not legally required, documenting the risk assessment and reasons for your decision can help demonstrate accountability.
What does the FCA expect from financial advice firms using AI?
The FCA does not prohibit financial advice firms from using generative AI.
It has also indicated that it intends to rely on existing regulatory frameworks rather than introducing a separate set of AI-specific rules.
These include the Consumer Duty, the Senior Managers and Certification Regime and existing expectations concerning governance and controls.
You can read more about this in the FCA’s approach to AI in financial services.
For advice firms, the practical implications include maintaining appropriate systems and controls, overseeing technology providers and ensuring that client communications remain accurate, understandable and suitable for their intended purpose.
The Consumer Duty is particularly relevant when AI contributes to information provided to retail customers.
A report that is generated quickly but contains misleading information or fails to explain an important risk is not improved simply because it was produced using new technology.
Professional oversight remains essential.
Do you need to tell clients that AI is being used?
There is no universal UK GDPR requirement to obtain separate consent every time AI assists with a document.
However, your firm must meet its transparency obligations and establish an appropriate lawful basis for the processing involved.
Whether a specific disclosure is necessary depends on the nature, purpose and circumstances of the AI use.
You should review your privacy information to determine whether it adequately explains the relevant processing activities and recipients.
This is particularly important if AI changes how client information is handled or introduces new third-party processors.
Where a provider processes personal data on your behalf, appropriate contractual arrangements under Article 28 of UK GDPR may also be required.
5. Practical Steps: Auditing Current AI Use and Setting a Firm-Wide Policy
The most effective starting point is to identify how AI is already being used across your firm. You can then evaluate the tools, information and workflows involved before deciding which activities should be approved, restricted or prohibited. A practical AI policy should give employees clear instructions rather than simply telling them to use technology responsibly.
You may find that AI adoption has developed gradually without anyone making a formal decision.
An adviser uses Copilot to improve an email. A paraplanner uses ChatGPT to summarise technical material. Someone else experiments with Claude when preparing meeting notes.
Individually, these activities may appear manageable.
Across the organisation, however, they can create different approaches to confidentiality, document handling and review.
Step 1: Understand what employees are already doing
Begin by speaking to the people involved in preparing advice documents.
Ask which tools they use, what activities they support and whether client information is being entered into those systems.
The purpose should be to understand the existing process, not discourage employees from identifying useful opportunities.
An audit might reveal that general-purpose AI is mainly being used for non-sensitive administrative tasks. Alternatively, it may uncover client-specific workflows that require immediate review.
Step 2: Categorise AI activities by risk
Not every AI task requires the same safeguards.
You could establish an internal framework similar to the following:
| AI activity | Example | Suggested approach |
| General administrative support | Drafting a non-sensitive internal meeting agenda | Permit through approved tools and standard controls |
| Technical research support | Explaining a general pension concept without client information | Allow with source verification and professional review |
| Client document summarisation | Extracting information from a fact find or meeting transcript | Require an approved platform and suitable data protection controls |
| Suitability report preparation | Producing client-specific advice documentation | Use a governed workflow with verified inputs and professional review |
| Automated recommendation decisions | Selecting or approving recommendations without meaningful professional oversight | Do not permit without specific legal, regulatory and risk assessment and appropriate safeguards |
This framework is illustrative rather than an FCA classification.
Your firm’s policy should reflect its actual activities, regulatory permissions and risk assessment.
Step 3: Establish clear rules for client information
Employees should understand what information can be shared with approved tools and which activities require additional authorisation.
Your policy should explain:
- Approved AI Platforms: Which Tools Employees May Use For Business Purposes.
- Permitted Information: What Types Of Client Data Can Be Processed And Under Which Conditions.
- Prohibited Activities: Which Uses Are Not Permitted Without Further Approval.
- Data Minimisation: How To Avoid Sharing More Information Than Necessary.
- Document Review: Who Must Check AI-Generated Suitability Reports And Other Client Communications.
- Incident Reporting: What Employees Must Do If Information Is Shared Incorrectly.
- Policy Ownership: Who Is Responsible For Updating Rules, Training And Approved Tools.
These instructions should be specific enough that employees can apply them during everyday work.
A policy that simply states ‘do not share confidential information’ is unlikely to answer the practical questions paraplanners face when preparing reports.
Step 4: Review technology providers before implementation
Once you have identified appropriate use cases, examine the providers capable of supporting them.
Look beyond the quality of the generated text.
Consider the provider’s security arrangements, processing locations, access controls, data retention policies, subprocessors and contractual commitments.
You should also evaluate whether the software allows your firm to maintain consistent report templates and approved wording.
For financial advice teams, paraplanning software designed around existing report production processes may provide a more manageable option than a collection of individual AI subscriptions.
Ammonite’s Planbot has been developed with financial planning documentation in mind. Its approach focuses on using your own templates and relevant source information to support document production while leaving the final content subject to professional review.
That can help your firm establish a more consistent process, although the suitability of any platform still needs to be considered against your own requirements.
Step 5: Review the policy as AI use develops
An AI policy should not be treated as a document that is written once and forgotten.
Providers introduce new features, employees discover different ways to use tools, and your firm’s reporting requirements may change.
A regular review should examine whether approved systems are being used as intended and whether any new processing activities require additional assessment.
Training is also important.
Employees need to understand not only which tools are permitted, but why particular controls exist.
For example, knowing that a client file should not be uploaded into an unapproved chatbot is useful. Understanding how to complete the task through an approved alternative is more helpful.
Can Financial Advisers Use ChatGPT With Client Data?
Financial advisers are not automatically prohibited from using ChatGPT with client data, but they must ensure that the specific service, account configuration and processing arrangements satisfy their legal and regulatory obligations. You should not upload identifiable client information through an unassessed personal account. Any approved use should have appropriate contractual, security, governance and professional review controls.
Whether an AI platform is suitable depends on the actual workflow.
Using a chatbot to improve a general explanation of investment volatility is different from uploading a complete client fact find.
The latter introduces personal data processing obligations that your firm must assess and manage.
An enterprise subscription may provide relevant safeguards, but approval should be based on the specific service and how it will be used.
Does AI Software Have to Store Client Data to Produce Suitability Reports?
No. AI-assisted document production does not necessarily require the software provider to retain client information after processing. However, information generally needs to be processed in order to produce the output, and the precise arrangements vary between providers. Your firm should distinguish between temporary processing, persistent storage, backups, logs and the separate question of whether information is used for model training.
For advice firms, reducing unnecessary retention can be an important part of a data minimisation strategy.
Ammonite’s Planbot is designed with this consideration in mind and states that sensitive client data used to populate reports is not stored or used to train AI models.
It is still appropriate to review the full processing arrangements before selecting any provider.
Protecting Client Data Without Losing the Benefits of AI
General-purpose AI tools can help financial advice firms work more efficiently, but client confidentiality must remain central to how the technology is introduced.
The main challenge is not simply deciding whether ChatGPT, Copilot or Claude is safe to use.
You need to understand which information is being processed, whether the provider’s arrangements are appropriate and how employees use the technology in practice.
A structured approach can help you introduce AI without creating unnecessary uncertainty around data protection.
That means assessing your current processes, selecting appropriate tools, establishing clear policies and maintaining professional oversight of client documentation.
It also means recognising that the most suitable technology is not always the most general-purpose option.
When your objective is to produce suitability reports and other financial planning documents, software designed around the actual requirements of the advice process may offer a better fit.
Explore a More Controlled Approach to AI Document Production
If your firm wants to reduce the time spent preparing suitability reports while maintaining control over its documentation and approach to client information, it is worth exploring technology developed specifically for financial advice.
Ammonite AI has created Planbot to help advisers and paraplanners produce reports using their existing templates, with control over wording, structure and presentation.
The platform is designed to support efficient document production without retaining the sensitive client data used to populate reports or using it to train AI models.
Book a Planbot demo to see how a structured AI reporting workflow could support your team while helping you maintain the standards expected of a professional financial advice firm.
