Skip to main content
Back to Articles

Is It Safe to Upload Your Bank Statement to an Online Converter? A 10-Point Security Checklist

Is it safe to upload a bank statement to an online converter? A 10-point security checklist covering PDF retention, encryption, AI subprocessors and deletion.

Save 10 hours monthly on data entry

Join accountants who automate their extraction

Try for free

Uploading a bank statement to an online converter is safe or unsafe depending mainly on one thing: whether the service keeps your source PDF after extraction, and for how long. Encryption is necessary but it is not the answer — TLS protects the file in transit and disk encryption protects it against a stolen server, yet neither stops a company from retaining, reading or being compelled to produce a document it chose to store. A converter that extracts your transactions and destroys the file within the same request exposes you for seconds; one that stores it indefinitely exposes you for as long as the company exists, plus every breach, legal request and change of ownership along the way. Before uploading anything, look for three things in writing: the retention period for the source file, the list of subprocessors that see it including AI providers, and the deletion procedure. If any of the three is missing or vague, you already have your answer.

Security checklist for uploading a bank statement PDF to an online converter

What your bank statement actually reveals

A bank statement is one of the densest identity documents you own: a single PDF links your name and address to your account number, your employer, your income, your creditors and your spending habits. That combination is worth far more to a fraudster than any single field on it.

Data on the statementWhat an attacker can do with it
Name and postal addressBuild an identity dossier; pass knowledge-based checks
IBAN or account numberSet up a direct debit (reversible), not withdraw funds
Bank name and branchMake a phishing call credible by naming your bank
Employer, salary, pay dateTime a scam for the day the money lands
Recurring payments and creditorsImpersonate a creditor and redirect a payment you owe
Counterparties (accountant, notary, supplier)Run business email compromise against you or against them

The realistic threat is not a criminal draining your account with your account number alone. It is impersonation: a scammer poses as a creditor you already deal with — a contractor, a notary, a landlord — and asks you to send your next payment to new bank details. Your statement makes that work because it names the creditor, the amount and the date. That is the pattern behind business email compromise, one of the costliest categories of reported fraud tracked by the FBI’s Internet Crime Complaint Center.

The legal classification is counter-intuitive. Financial data is not a special category of personal data under GDPR Article 9, and under the CCPA an account number becomes sensitive personal information only when combined with a security code, password or credentials. A statement is ordinary personal data in the legal sense — which is why some services handle it more casually than it deserves.

The one question that matters most: does the service keep your PDF?

The single most useful question to ask any online bank statement converter is whether it keeps the source PDF after extraction and for how long, because encryption answers a different question. Retention is a business decision; encryption is a technical control.

Encryption in transit protects the journey: TLS means nobody on the network can read the file as it passes, but at the destination it is decrypted and exists in the readable form it had on your computer. Encryption at rest protects the medium — a stolen disk, a lost backup, an open bucket. Neither defends against an employee with legitimate access, a lawful production order, a subprocessor holding a copy, or an acquisition in which the database is part of the assets sold.

The law is clear on the principle and silent on the number. GDPR Article 5(1)(e) requires personal data to be kept in identifiable form no longer than necessary, and Article 17(1)(a) gives a right to erasure once it is no longer needed; neither states a duration. The closest thing to a ceiling comes from US law: the FTC Safeguards Rule requires covered financial institutions to encrypt customer information in transit and at rest, and to dispose of it securely no later than two years after its last use. Whether a general-purpose converter falls inside that definition depends on what it does, so treat it as a benchmark rather than an obligation: a service that publishes no duration at all is not even meeting that minimum. Verify the rules in force in your jurisdiction.

The 10-point security checklist

Ten questions are enough to classify any online bank statement converter, covering source-file retention, encryption, staff access, AI subprocessors, hosting jurisdiction, deletion, account security, in-memory versus on-disk processing, retention of the extracted transactions and the quality of the policy itself. Each should be answerable from the service’s public documentation, without having to email support.

The question to askGood answerRed flag
1. Is the source PDF kept, and how long?Destroyed when the request ends; retention stated in hours”As needed”; no duration anywhere
2. Is data encrypted in transit and at rest?TLS plus encrypted storage and backups, listed separately”Bank-level encryption”; HTTPS as the whole answer
3. Who at the provider can read my file?Role-based, logged access; support cannot open it unrecordedSilence; support asks you to email it
4. Which AI subprocessors see it, and what do they keep?Providers named, paid tier, retention window, training excludedNo list; “we use AI” and nothing more
5. Where is the data hosted?Region and jurisdiction named; transfer mechanism statedUnknown, or “worldwide”
6. Is deletion real, and does it cover backups?Immediate on live systems; rotation window stated”Deleted regularly”; deletion that only hides the item
7. How is my account protected?Password rules, two-factor authentication, session expiryNo 2FA on an account holding financial data
8. Processed in memory or written to disk?In memory, or a temp file removed with the requestUploads dropped into object storage by default
9. How long are the extracted transactions kept?While your account exists; deletable one by oneRetention stated for the PDF only
10. Is the policy public, dated and specific?Public URL, last-updated date, durations as numbersUndated, template text, contradictions between pages

Point 4 is the one most often left undocumented. GDPR Article 28 forbids a processor from engaging another processor without prior written authorisation, which is why a serious service publishes its subprocessor list. If a converter uses AI to read your statement, that provider sees your document and its retention policy becomes yours.

AI provider behind the converterAre your documents used to train models?How long the provider keeps the inputThe clause that catches people out
OpenAI (API)No, unless the customer explicitly opts inAbuse-monitoring logs kept up to 30 days by default, and those logs can contain prompts and responsesZero data retention is granted only to approved customers; it is not a setting a converter can simply tick
Anthropic (commercial API)No, not by defaultInputs and outputs deleted within 30 daysSessions flagged by automated trust and safety systems can be kept up to two years even under a zero-retention arrangement, and a document sent through the Files API stays until explicitly deleted
Google Gemini (unpaid tier)Yes — submitted content and generated responses are used to improve and develop Google productsNo fixed window published for unpaid servicesHuman reviewers may read, annotate and process API input and output; paid services exclude product improvement, but Search grounding stores prompts and output for 30 days regardless

The consequence is blunt: a converter running on an unpaid AI tier may be operating under terms that permit human review of your bank statement. These terms are revised regularly, so ask which provider and which tier a service uses, read that provider’s current policy yourself, and distrust any claim that zero retention was simply switched on.

Point 6 is where a policy can be technically true and practically misleading. Deleting a row from a production database does not remove it from yesterday’s backup, so “your file was deleted immediately” and “a copy exists in our backups for thirty days” can both be true. The useful question is not “do you delete?” but “what is the backup rotation window?”

Point 8 is decisive. A converter that holds the PDF in memory has nothing to purge; one that writes it to an object store now has the file in a bucket, in logs, and in every backup.

Bank statement PDF deleted after extraction versus the same file stored in a database

Red flags that should make you close the tab

Six signals disqualify an online bank statement converter before the ten-point checklist is worth running at all, because each one shows the operator has not thought about the document it is asking you to upload.

  • No privacy policy, or one that never mentions uploaded files.
  • No HTTPS. Non-negotiable — but the reverse proves nothing: certificates are issued automatically to anyone who controls a domain name, and the vast majority of phishing sites now serve their pages over HTTPS. The padlock shows the connection is encrypted, not that the operator is honest.
  • “Your files are deleted regularly” with no interval. Regularly is not a duration.
  • A browser extension asking to read and change data on all websites. A converter has no business in your banking session.
  • Any service asking for your online banking credentials. Never. Reading a PDF you already downloaded requires no bank access.
  • No legal identity — no company name, no address, no data protection contact — so no way to exercise an erasure right.

How to reduce your exposure before you upload anything

Three things about a bank statement upload stay under your control whatever the converter’s policy says: the content you send, the account you send it from, and how long the file survives on both sides. The seven steps below reduce all three.

  1. Check whether your bank already exports the format you need. Most portals offer CSV, OFX or QIF for recent months, and a native export beats any OCR pass.
  2. Send only the months you need. GDPR Article 5(1)(c) requires personal data to be adequate, relevant and limited to what is necessary — a decent personal rule too.
  3. Redact by deleting, not by covering. A black rectangle drawn in a PDF viewer is a separate object sitting on top of text that remains intact in the content stream: select and copy the page into a text editor, or delete the annotation layer, and the hidden characters come straight back. What works: remove the text in the source document and re-export to PDF, use a redaction tool that deletes the underlying content, or print, mask and rescan the page.
  4. Check the document properties. Converting a document to PDF does not strip metadata; author names, file paths and revision history survive.
  5. Use a dedicated account with a strong unique password and two-factor authentication. If the service holds your extracted transactions, that password is the last control between credential stuffing and your financial history.
  6. Verify the output before deleting anything. Check opening balance, closing balance, transaction count, date range and a sample of amounts against the PDF.
  7. Delete the source file, then the extracted data once you no longer need it. A converter is a processing step, not an archive.

One point deserves its own warning: if your bank encrypts the statement, remove that protection yourself before uploading rather than looking for a converter willing to take the password. No legitimate service needs your bank’s PDF password, and handing one over adds a credential to the list of things a breach would expose — our guide to dealing with an encrypted statement PDF covers the safe ways to open it.

What “we never store your PDF” should mean in practice

The claim “we never store your PDF” only means something if the service states where the file lives during processing, when it is destroyed, and what survives. Here is a definition precise enough to be audited.

  1. The file exists only for the duration of the extraction — in memory or in a temporary file — and is destroyed as soon as the extraction ends. The retention window is written as a number rather than left as “as needed”.
  2. It is never copied elsewhere: not into object storage, not onto a support ticket, not kept for debugging, not mirrored into logs.
  3. Only the extracted transactions persist, attached to your account: the PDF is the input, not the product.
  4. Deletion is real: deleting an extraction removes the content instead of setting a hidden flag.
  5. The same rule applies to the API, which is another door into the same pipeline.
  6. No endpoint returns your original file. If the file is not kept there is nothing to serve, and that absence is the strongest available proof the policy is real.

Measured against that definition, this is what BankStatementLab does:

  • The source PDF is deleted from disk the moment extraction succeeds, and the database reference to that file is cleared in the same operation.
  • No endpoint returns the original file, in the application or in the public API: the capability does not exist.
  • The temporary files created when the statement is split into pages are cleaned up automatically.
  • Nightly purges: extractions made without an account are removed after 24 hours.
  • Automatic deletion is yours to set, from 1 to 30 days — off by default, starting at 14 days when you switch it on, applied hourly to your own value.
  • Deleting is deleting: removing an extraction clears its transaction data too, and deleting your account erases every extraction and file for real.
  • File deletions are protected against directory traversal, so a crafted path cannot reach outside its folder.

Two caveats belong in the same breath: the file is written to disk during extraction rather than held in memory, so the deletion step is what does the work; and point 2 of the definition — never copied elsewhere, never mirrored into logs — is something to ask any provider in writing rather than take from an article. Treat the list as a definition, not as something to accept on faith, and run the ten questions above against any service, including this one.

The source document and the extracted data are two different questions

The source PDF and the extracted transactions raise two separate retention questions, and most retention claims collapse because they are mixed. The source document is the input: a PDF a converter has no reason to hold once it has been read. The extracted data is the output — the transactions you came for — worthless if it vanishes before you export it. “We store nothing” is therefore either untrue or useless.

That separation is the one implemented above. The PDF goes as described; the transactions stay until you delete them, or until your own rule deletes them for you at any delay from 1 to 30 days. That is the concrete answer to “how long does my data stay?”: until you delete them, or for the number of days you set once you switch that rule on.

How BankStatementLab answers the 10 questions

The ten-question checklist above, applied to BankStatementLab. Four of the ten are answered by the product itself; the rest are questions you should put in writing to any provider, including this one, and they are marked as such rather than padded with reassurance.

The question to askBankStatementLab
1. Is the source PDF kept, and how long?Deleted as soon as extraction succeeds, database reference cleared with it.
2. Encrypted in transit and at rest?Traffic runs over HTTPS, which covers transit only. Encryption at rest is a separate control and should be asked for in writing, here as with any provider — a padlock never answers it.
3. Who at the provider can read my file?Not answered by this article: put the question in writing to any provider, including this one, and expect an answer naming roles and logging, not reassurance.
4. Which AI subprocessors see it?Pages are read by the vision capabilities of Anthropic, OpenAI or Google Gemini models. Ask which tier and which retention window applies before uploading a client’s statement.
5. Where is the data hosted?Not answered by this article: ask for the hosting region, the jurisdiction and the transfer mechanism, and treat “worldwide” as a non-answer.
6. Is deletion real, and does it cover backups?Deletion is real on live systems: an extraction is removed with its transaction data, deleting an account erases every extraction and file, and file deletions are guarded against directory traversal. The backup rotation window remains a question to ask.
7. How is my account protected?Ask for the specifics — password rules, two-factor authentication, session expiry — rather than accepting “secure by design” from any service.
8. Processed in memory or written to disk?Written to disk during extraction, then deleted the moment it succeeds; the temporary files created when the statement is split into pages are cleaned up automatically.
9. How long are extracted transactions kept?Until you delete them, or until your own automatic-deletion rule fires: any delay from 1 to 30 days, off by default, 14 days when you switch it on.
10. Is the policy public, dated and specific?Read the published policy and check two things — a last-updated date and durations written as numbers — here as anywhere else.

What to do if you already uploaded a statement somewhere you now doubt

If you have already uploaded a bank statement to a service you no longer trust, the useful actions are erasure, credential hygiene and account monitoring, in that order.

  1. Delete the file and the extracted data inside the product, and keep a dated screenshot of the confirmation.
  2. Send a written erasure request. GDPR Article 17 covers data no longer necessary for the purpose it was collected for. Ask for three confirmations: deletion from live systems, the backup rotation window after which copies disappear, and deletion by subprocessors, whom Article 28 requires to be bound by the same data protection obligations as the processor itself.
  3. Change that password, and anywhere you reused it. Secure your email account first with two-factor authentication: it is the recovery channel for everything else.
  4. Watch the account. Read the next few statements line by line and switch on transaction notifications.
  5. Handle unauthorised direct debits through your bank. EU payment rules give a payer eight weeks from the debit date to have an authorised SEPA direct debit reversed, without having to justify the request, and the amount is credited back with the value date of the original debit; for a transaction you never authorised at all, the deadline to notify your bank runs up to 13 months. Verify the deadlines in force with your bank.
  6. Ask your bank about account-level controls: blocking creditors, an allow list of mandates, or new account details.
  7. Treat unexpected contact as hostile. Someone holding your statement can name your bank, your employer and your last payment. Call back on the number printed on your card.
  8. Expect to be told if there is a breach. GDPR Article 33 requires a controller to notify the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of a personal data breach.

Key takeaways

  • Safety depends on source-file retention, not on encryption: TLS protects the transfer and disk encryption protects the medium, but neither stops a company from keeping, reading or being compelled to produce a file it decided to store.
  • A bank statement is not a special category under GDPR Article 9, and an account number alone is not sensitive personal information under the CCPA — the legal classification is weaker than the real risk, impersonation of the bank, employer and creditors named on the page.
  • If a converter uses AI, that provider’s terms become yours: OpenAI keeps abuse monitoring logs up to 30 days by default, Anthropic can retain flagged sessions up to two years even under a zero-retention arrangement, and Google’s unpaid Gemini tier permits human review of API input and output.
  • “We delete your file immediately” and “a copy exists in our backups for another month” can both be true, so the real question is the backup rotation window.
  • The strongest evidence that a service does not keep your PDF is the absence of any endpoint able to return it.
  • Before uploading: prefer your bank’s native CSV or OFX export, send only the months you need, redact by removing text rather than covering it, then delete the source file.

Frequently Asked Questions

Is it safe to upload a bank statement to an online converter?

It depends almost entirely on whether the service keeps your source PDF after extraction, and for how long. A converter that destroys the file at the end of the request exposes you for seconds; one that stores it exposes you for as long as the company exists. Check its retention policy, subprocessor list and deletion procedure first.

Can someone empty my bank account with just my IBAN?

No. An account number alone does not allow anyone to withdraw funds, and an unauthorised direct debit can be reversed through your bank — EU payment rules give eight weeks from the debit date, and up to 13 months to report a transaction you never authorised. The real danger is impersonation, since a statement names your bank, employer and creditors.

Does the HTTPS padlock mean my bank statement is safe?

No. The padlock proves the connection is encrypted; it proves nothing about the operator’s honesty, its retention policy or its subprocessors. The vast majority of phishing sites serve their pages over HTTPS as well, so treat the padlock as a minimum requirement, not as evidence of safety.

Do online converters send my bank statement to AI providers like OpenAI, Anthropic or Google?

Many do, and those providers’ retention terms then become part of the service’s. GDPR Article 28 forbids a processor from engaging another without prior written authorisation, which is why a serious provider publishes a subprocessor list. If a converter will not name its AI providers, you cannot assess what happens to your document.

Is a bank statement classified as sensitive data under GDPR?

No, not in the legal sense. GDPR Article 9 special categories cover racial or ethnic origin, political opinions, religious beliefs, trade union membership, genetic and biometric data, health, sex life and sexual orientation — financial data is not on that list. Under the CCPA, an account number qualifies only when combined with a security code, password or credentials.

How long is an online converter allowed to keep my bank statement?

There is no fixed maximum in EU law. GDPR Article 5(1)(e) requires storage for no longer than necessary and Article 17 erasure without undue delay, without ever naming a duration. The clearest benchmark is the US FTC Safeguards Rule, which makes covered financial institutions dispose of customer information no later than two years after its last use. In practice, a converter has no reason to keep a source PDF beyond the request that reads it. Verify the rules in force in your jurisdiction.

Should I ever give an online converter my online banking login?

Never. Converting a PDF you already downloaded requires no access to your bank account, so a request for banking credentials has no legitimate justification. If you have already entered them, change your banking password and contact your bank.

Does BankStatementLab store my bank statement after conversion?

No. The source PDF is deleted from disk as soon as the extraction succeeds, its database reference is cleared with it, and no endpoint in the application or the public API can return the original file.

Can I make my extracted data delete itself automatically?

Yes. Automatic deletion can be set to any delay from 1 to 30 days; it is off by default, starts at 14 days when enabled, and an hourly job applies your own setting. Deleting an extraction yourself also clears its transaction data.


If you want a converter that treats the source PDF as a transient input rather than an asset, that is the model BankStatementLab is built on: the statement is read, the transactions are extracted into Excel or CSV, and the PDF is deleted as soon as the extraction succeeds — in the app and through the API alike. No tool reads a bank statement without ever getting something wrong, so check the extracted lines against the PDF whatever produced them. Create an account at https://app.bankstatementlab.com/signup and run the ten questions above against it.

---
🎁 5 credits on signup, then 5/month
💎 1 credit = 1 page

Ready to Automate your accounting?

Join thousands of professionals who save hours every month.

Try BankStatementLab
Written by bankStatementLab Team