Is It Safe to Upload Bank Statements to ChatGPT?
Is it safe to upload bank statements to ChatGPT? What a statement reveals, what OpenAI keeps, ChatGPT Finances and safer ways to analyse one.
- Published
- Updated
No, a personal ChatGPT account is not a safe default for uploading a client’s unredacted bank statements. Consumer accounts can retain the upload and use eligible content for training unless you change the settings. Turning training off leaves confidentiality, access and retention questions to resolve.
Is It Safe to Upload Bank Statements to ChatGPT?
Uploading your own bank statements is a personal decision about disclosing your financial life to another provider. Uploading a client’s statements is a professional decision involving information entrusted to your firm. A client’s permission to analyse statements does not automatically settle every disclosure to an AI provider.
The Office of the Australian Information Commissioner (OAIC) recommends that organisations keep personal information, particularly sensitive information, out of publicly available generative AI tools. Its guidance on commercially available AI products explains the privacy assessment organisations need to make.
For firms covered by the Australian Privacy Principles, the purpose of the disclosure and any applicable exception need assessment. Consider information about joint account holders and transfer recipients as well as your client. Lawyers also need to protect confidential communications and assess privilege in the relevant matter.
What a Bank Statement Gives Away
A bank statement can identify the account holder and reveal financial relationships far beyond the closing balance. Depending on the statement, it can expose:
- The account number and BSB, together with the holder’s name and address.
- An employer’s name, salary amounts and payment frequency.
- Spending patterns and the places a person visits.
- Payments to health providers or legal advisers.
- Other people’s names and payment references in transfers.
If someone obtains those details through an account compromise or provider breach, they can make an impersonation attempt more convincing. Knowing a person’s bank, employer and recent payment gives a scammer material for a targeted message. Account details alone do not establish that someone can withdraw money.
In a fictional example, Jordan’s statement shows a regular wage and a payment to a named solicitor. An attacker could impersonate that solicitor and refer to the payment when seeking another transfer. The Scamwatch document-sharing checklist identifies sending bank statements to someone as a warning sign to investigate.
Removing the header alone leaves the transaction descriptions and relationships visible. Redaction needs to cover information that identifies people within the transactions too.
What OpenAI Keeps From an Upload
OpenAI’s treatment of an uploaded PDF depends on the workspace and data controls. Paying for Plus does not give a personal account the same data terms as Business or Enterprise.
| Plan | Model training | Retention | Human access |
|---|---|---|---|
| Free | Eligible personal-workspace content can be used by default. Turn off Improve the model for everyone to opt out. | Saved chats remain until deleted or removed under an applicable policy. Saved files can remain separately. | Authorised personnel and service providers can access content for specified purposes. |
| Plus | The same personal-workspace training control applies. | Chat and saved-file deletion are separate actions. | The consumer access rules apply. |
| Business | Business content is excluded from training by default, unless explicitly shared for that purpose. | Workspace retention and deletion terms apply. | OpenAI limits employee access to support, abuse investigation and legal compliance. Bound contractors can review abuse or misuse. |
| Enterprise | Business content is excluded from training by default, unless explicitly shared for that purpose. | Administrators control workspace retention. | Authorised OpenAI employees can access conversations for incident resolution, permission-based recovery or legal requirements. |
OpenAI’s business privacy commitments describe the Business and Enterprise arrangements. Training exclusions are separate from permitted access.
For consumer accounts, authorised human access can cover security incidents, support and legal matters. Model-improvement access depends on the opt-out. OpenAI explains these purposes in how it handles consumer data.
Deleting a saved chat removes it from view immediately and normally schedules permanent deletion within 30 days. Exceptions include security or legal retention and content already de-identified and disassociated from the account. Files saved in Library or Space require separate deletion. Enterprise saved files follow workspace policy, and backups can persist after files become inactive. OpenAI’s chat and file retention rules explain those distinctions.
ChatGPT Finances and Linked Bank Accounts
ChatGPT Finances connects supported financial accounts for personal-finance questions. OpenAI lists it for eligible users in the United States. Linking accounts adds an automatically syncing financial-data connection to the information you put in a chat.
It can read available balances and transactions, with other fields depending on the institution. Optional credit-report data is a separate connection. The dashboard helps explain spending and recurring bills. ChatGPT cannot move money or pay bills through Finances.
Finance conversations follow your ChatGPT training settings. Financial memories can retain details you share about goals or obligations for later conversations.
To disconnect, open Settings > Apps > Finances and remove the bank connection. OpenAI says synced account data is deleted from its systems within 30 days after disconnection. Delete relevant conversations and financial memories separately. A credit-report connection also needs separate removal. The Finances help page explains connection controls and provider-specific deletion terms.
If a Client Has Already Uploaded Statements
Establish what the client shared before deciding how to contain the disclosure. For an active matter, consider evidence-preservation duties before deleting material that records what happened.
- Record the uploaded filenames, account or workspace used and the questions asked. Note whether the client included legal advice, annotations or another person’s records.
- Subject to preservation requirements, delete the relevant chats and separately saved PDFs. Check Library or Space, project files and shared links. Archiving a chat leaves it stored.
- Review Settings > Data controls and turn off Improve the model for everyone for future eligible personal chats. A settings change does not remove existing chats. OpenAI’s data controls instructions explain the available controls.
- Check saved memories and any linked financial accounts. Remove unwanted stored information and connections separately, then record the actions taken in the matter file.
Confirm that the relevant chats and files no longer appear in the account. Keep the provider’s deletion period separate from that visible result. The guide to deleting ChatGPT history covers the deletion process in more detail.
An upload needs closer legal assessment when it includes advice from the lawyer, privileged annotations or documents subject to disclosure restrictions. Statements about another party also raise confidentiality and privacy questions. The Law Society of NSW’s IT guidance explains the care required for electronic client communications. Have the lawyer assess any effect on privilege under the rules applying to the matter.
Safer Ways to Analyse Statements
Choose the handling arrangement before sending the records. Each route changes who receives the information and what your firm must control.
A business AI workspace excludes content from model training by default and gives the firm an organisational environment. Assess its access, retention and contract terms for the intended work. The subscription alone does not approve a client disclosure.
Redacting a working copy reduces the information sent to the provider. Keep the original unchanged, remove unnecessary identifiers and check the exported copy for recoverable text. Detailed transaction patterns can still identify a person after names disappear.
A model running entirely on your firm’s own equipment can keep processing within that environment. Confirm that the application sends no prompts or files to a cloud service. Your firm still controls device security and access to saved outputs.
A specialist service lets the firm agree confidentiality obligations for the work. FinTrace signs confidentiality deeds on request and delivers statement analysis as a service. Agree the deed and document-handling terms before transferring client records. Choose the route your firm can authorise for this matter, then send only the information the task requires.
FinTrace
Statements in. Evidence out.
FinTrace turns your bank statements into one searchable Excel ledger and a written findings report, with every finding cited to its source page.
FinTrace
See how FinTrace works