AI data leakage in insurance broking: what leaves when client files go into ChatGPT or Claude.
An insurance broker’s file is a map of everything that could go wrong for a client: what they own, what it is worth, what has been claimed before, who has been injured, what the insurer must be told. The Insurance Brokers Code of Practice commits subscribers to act in the client’s best interests, and the Privacy Act treats the health information in a claims file as sensitive information. This series takes seventeen documents a brokerage produces every week, from the renewal recommendation to the complaint response, marks every item the brokerage is obliged to protect, and shows where each one goes when the document is pasted into ChatGPT or Claude. Each page ends with the same document on Ermos, inside the brokerage.
The documents, one at a time
Sums insured, premiums, the claims history, the risk address, the security and the advice. The client’s whole risk picture, written in a hurry.
See what leaves → Report or memoThe claims chronologyThe injured person, the incident, the witnesses, the medical reports and the reserve. A claim file, in order.
See what leaves → Report or memoThe policy comparisonWhich insurers quoted, what they charged, what they excluded, what they pay the brokerage, and the risk they priced. The market, mapped for one client.
See what leaves → EmailThe submission to an underwriterFinancials, directors, claims, security, contracts and the pitch. Everything the underwriter needs, which is everything.
See what leaves → EmailThe claims email to a clientThe insurer’s position, the indemnity problem, a lapsed policy, the director’s personal exposure and the advice. The email nobody wants to write.
See what leaves → FormThe needs analysisRevenue, payroll, properties, contracts, the directors and the gaps in cover. The business, learned in one meeting.
See what leaves → FormThe renewal declarationClaims, convictions, insolvency, cancelled policies and changes in the risk. The client’s admissions, on a form the law requires to be complete.
See what leaves → LetterThe personal advice letterThe home, its contents, an investment property, the cars, the family, the prior claims and the recommendation. A household’s insurable life.
See what leaves → EmailThe premium funding applicationBank details, directors’ identities, premiums and a payment schedule. A finance application, drafted as a routine email.
See what leaves → LetterThe complaint responseThe allegation, the policy wording, the broker’s file, the insurer’s position and the offer, in a letter AFCA may read next.
See what leaves → SpreadsheetThe claims registerEvery client’s open claims, reserves, injuries and insurer positions in one grid, with the brokerage’s own notes beside them.
See what leaves → SpreadsheetThe asset scheduleSerial numbers, rooms, sums insured and where the keys are kept, in the grid prepared for every renewal.
See what leaves → FormThe cyber insurance questionnaireRecord counts, MFA gaps, backup weaknesses, payment controls and recent losses, written down for an insurer.
See what leaves → LetterThe claim denial dispute letterThe insurer’s reasons, the client’s weakest facts, lost income and the demand, in a letter written under pressure.
See what leaves → Report or memoThe loss adjuster reportThe interview, the technician’s findings, the coverage comments and the figures, written for the insurer and read by the broker.
See what leaves → LetterThe broker transfer letterThe trust, the property, the premium, the claims history and the client’s reasons for leaving its old broker.
See what leaves → EmailThe certificate of currency requestA new contract, the cover the head contractor demands, a lapse in workers compensation and what the client must disclose.
See what leaves →How the leakage happens
The mechanism is the same for every document. The whole text is transmitted to the provider’s servers. On personal plans the user cannot choose where it is stored or processed, and for the major providers that has generally meant outside Australia. On consumer tiers it can be retained and used to train future models. The provider’s staff and systems can reach it under the provider’s own policies. Once sent, the brokerage no longer controls it. Our on-premise vs cloud comparison sets out the four ways to run AI and where each one puts your documents.
The obligations in play
Each marked item on these pages links to the guide that covers its obligation:
- The Privacy Act, the APPs and AI
- APP 8 cross-border disclosure and AI
- Choosing an on-premise or private cloud AI provider
The same documents on Ermos
Ermos Dominion drafts and searches on a unit in your office, so the documents never leave it. Ermos Edge keeps your documents on your own instance in an Australian data centre and sends only permitted excerpts, per question, to a named Australian provider. Every page in this series ends with both, item by item.
Common questions
Can insurance brokers use ChatGPT or Claude with client information?
Do ChatGPT and Claude train on what staff paste in?
Do business plans such as ChatGPT Business or Claude Team solve the problem for insurance brokers?
What is the safest way for a brokerage to use AI with client documents?
What is AI data leakage in an insurance brokerage?
Which broking documents carry the most risk?
Does the Code of Practice say anything about AI?
See Dominion answering from your own documents.
A demonstration takes 30 minutes, uses no client data, and comes with no obligation.
Book a demo