Broker call transcripts and AI data leakage: what a recorded client call gives away in ChatGPT or Claude.

Many brokerages now record client calls and transcribe them so that the file shows what was said. A transcript is the most candid document in the file: clients talk about their jobs, their nerves, their families and their money in their own words, and the broker talks through the advice. Staff paste transcripts into ChatGPT or Claude to produce file notes and action lists. This page marks every item in a routine broker and client call transcript that the National Consumer Credit Protection Act and the Privacy Act bear on, and shows how the same transcript becomes a file note on Ermos inside the brokerage.

The broker and client call transcript

Produced after a recorded client call. ASIC’s RG 273.165 expects records of relevant conversations with the consumer and RG 273.167 lists audio recordings and file notes of conversations among the forms of record, so the recording and the transcript can be part of the evidence that the best interests duty was met.

An invented transcript. The speakers are fictional. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the brokerage

1. 1
   The speakers

   Three named people, recorded. Everything said below is attributed to them. Privacy Act guide.

2. 2
   Income in the clients’ words

   Base pay, overtime and a spouse’s salary, spoken aloud. Financial information the broker asks about because section 117 requires reasonable inquiries into the consumer’s financial situation. Privacy Act guide.

3. 3
   A job worry about a named employer

   A client’s fear about his employer’s tender and about colleagues who left. Said in confidence to get the loan right, and damaging if repeated. Privacy Act guide.

4. 4
   A late payment and the explanation

   The credit history point and the client’s account of it. Credit information, which Part IIIA of the Privacy Act regulates when it comes from a credit report. Privacy Act guide.

5. 5
   A family gift and a request for secrecy

   The donor, the amount, where it came from and an instruction not to contact a relative. A third person’s finances and a family confidence. Privacy Act guide.

6. 6
   A private family plan

   A statement about trying for a child and the working hours that would follow, and a spouse’s wish to keep it off the form. Family plans of this kind are private personal information. They would be health information only if they disclosed something about a person’s health or a health service. Patient data guide.

7. 7
   The spoken consents

   The broker’s request, and the clients’ answers, about recording and credit checks. The evidence of what the clients agreed to, which does not on its face reach an AI provider. Privacy Act guide.

8. 8
   The broker’s advice and a promise

   The statement about lender policy, the card closure and a commitment to check policy before lodging. The advice that section 158LA covers, and the sentence a later complaint may turn on.

9. 9
   Identity confirmations spoken aloud

   A date of birth and a home address read into the record. Identity verification data, now in a transcript. Privacy Act guide.

What happens in Frontier Models (ChatGPT, Claude)

• The whole document 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, under that country’s legal process. Some business plans offer storage in Australia, but the provider still receives the content.
• On consumer tiers, prompts can be retained and used to train future models.
• The provider’s staff and systems can reach the content under the provider’s own policies.
• Once sent, the brokerage no longer controls it. Retention and deletion happen on the provider’s terms.

The same document on Ermos

Part of AI data leakage in mortgage brokerages. See how this works in practice: Best interests notes on Ermos.

Common questions