All articles
Guide6 min read

AI meeting notes without exposing confidential data

Having AI summarize a meeting hands it names, figures and decisions. How to anonymize a transcript before ChatGPT, Claude or Gemini without losing meaning.

By Pierre de ONYRI

Having AI summarize a meeting saves real time — but a raw transcript is a concentrate of confidential material: speaker names, clients mentioned, amounts, strategic decisions. Before asking ChatGPT, Claude or Gemini for a summary or a decision log, replace these with tokens. The model returns the structure and the trade-offs; no identity or sensitive figure leaves the browser.

A raw transcript is a concentrate of confidential material

An automatic transcription captures everything, unsorted: who said what, which amounts circulated, which client and project names came up. Pasting it into an assistant to summarize it means handing the whole meeting to a third party — often outside your perimeter.

  • Speakers: names, roles, affiliations.
  • Third parties mentioned: clients, partners, candidates, competitors.
  • Figures: budgets, prices, targets, salaries mentioned.
  • Strategy: decisions, deadlines, internal project names.
Diagram: a meeting and its microphone feeding a transcript whose speaker names are replaced by tokens, before passing through an anonymization gate.
Speakers, amounts and decisions become tokens before sending; the thread of trade-offs stays readable.

What to anonymize in a transcript

  1. 1Names and roles of speakers and third parties mentioned.
  2. 2Names of clients, projects and entities cited.
  3. 3Amounts, prices and quantified targets.
  4. 4Contact details and identifiers slipped in verbally (emails, numbers).

A flow that keeps the confidential in-house

  1. 1Detection: the engine spots speakers, third parties, amounts and strategic markers.
  2. 2Tokenization: each element becomes a neutral token, kept in local memory.
  3. 3Sending: only the anonymized transcript goes to the AI — the meeting doesn't transit in the clear.
  4. 4Restoration: the summary is de-tokenized in your browser, ready to share internally.

ONYRI Sanitize detects a transcript's sensitive elements — names, entities, amounts, decision markers — and restores the answer in your browser. Your teams get usable meeting notes without ever exposing what was said.

Frequently asked questions

Can I have ChatGPT summarize a meeting?
Yes, as long as you don't hand it the transcript in the clear. Anonymize the speakers, the third parties mentioned and the figures before sending: AI produces a summary and a decision log, and you restore it all in your browser.
Are automatic transcription tools safe?
They capture everything unsorted, then send the content to a third-party service. The risk is in what leaves. The defense is the same: anonymize the transcript before an external model processes it, and keep the mapping local.
Is the summary still useful once names are masked?
Yes. Tokens are consistent, so the thread of exchanges and decisions is preserved. After restoration in your browser, you get an accurate, named summary, without identities ever leaving your machine.

Sources & references

Keep your sensitive data in your browser

ONYRI Sanitize detects and masks your sensitive data before it reaches the AI, then restores the answer — from names to API keys.

Anonymize my prompt

Read next