# Interpreter Confidentiality and AI Tools: What to Check First

> Interpreters remain bound by confidentiality when they use AI tools, so before using one on an assignment check where audio and text go, whether they are stored or used to train models, who can access them, and whether the client and your agency permit the tool. Disclose when required, and avoid general chatbots for confidential content. Unicaption is end-to-end encrypted, never stores audio, deletes transcripts when the session ends, and never uses content to train models.

Published 2026-09-30 · The Unicaption Team · Human-readable version: https://unicaption.com/blog/interpreter-confidentiality-ai-tools.html

## What your duty actually covers

Professional codes of ethics for interpreters, including those from major associations and certification bodies, consistently treat confidentiality as a core obligation. Interpreters don't disclose information learned during an assignment, and they protect it from being exposed through carelessness as well as intent.

That second part is where AI tools come in. If a recording, transcript, or document leaves your control because a tool uploaded it, retained it, or trained on it, that can be a breach even if you meant no harm. Your agency contract, the client's rules, and local law may also impose their own requirements on top of the code.

> This article is practical orientation, not legal advice. Read your own code of ethics, your contracts, and the rules that apply to your setting, and ask your agency or an attorney when you're unsure.

## Six questions to ask before using any tool on a job

**1. Where does the audio or text go?**
Does it stay on your device, or is it processed on a vendor's servers? Which countries? Is any of it passed to subprocessors? Vague answers are an answer.

**2. Is anything stored, and for how long?**
Recordings, transcripts, and logs can all persist. Look for a plain statement of what is kept and when it is deleted, not just "we take privacy seriously".

**3. Is your content used to train models?**
Consumer chatbot settings differ by product and plan, and they change. Check the current terms. If content can be used for training, don't put confidential material in.

**4. Who at the vendor can access it?**
Ask whether staff can read sessions, for support or quality review, and under what controls.

**5. Is it encrypted in transit and at rest?**
Encryption doesn't solve everything, but its absence settles the question.

**6. Do your client and agency allow it?**
The tool can be technically safe and still be prohibited on a given job. Many courts, clinics, and corporate clients have explicit rules on recording, transcription, and AI. When in doubt, ask first.

## Situations that need extra care

- ⚠ **Pasting assignment documents into a general chatbot.** Contracts, case files, and patient information shouldn't go into a tool whose retention and training terms you haven't confirmed.
- ⚠ **Recording a session "just for prep".** A recording is stored data, and laws on recording conversations vary by place and require consent in many.
- ⚠ **Notetaker bots joining a call.** They appear as participants, their output is stored by a third party, and many clients prohibit them.
- ⚠ **Free tools with unclear business models.** If you can't tell how it makes money, assume your data is part of it.
- ⚠ **Court, immigration, and clinical settings.** Rules are strictest here. Get explicit permission before using any tool.

## Disclosure: what to tell the client

Even when a tool is safe, transparency protects you. A short sentence in your confirmation, such as "I use a private live-captioning tool as a reference during remote sessions; nothing is recorded or stored", lets the client object before the job, not during it. If the client says no, the answer is no.

Be honest about what the tool does and doesn't do. A captioning reference that shows you the words is different in kind from a tool that records and summarizes the conversation. Name the difference when you describe it.

## How common tool types compare

|  | General chatbot | Meeting notetaker bot | Unicaption |
| --- | --- | --- | --- |
| Joins the call | No | Yes, as a participant | No, runs on your own device |
| Typical data kept | Depends on product and plan; check current terms | Often recordings and transcripts | Nothing: audio never stored, transcripts deleted at session end |
| Used to train models | Varies by product and settings | Varies | Never |
| Best use | Public-information prep | Internal meetings you own | Live reference during assignments |

## Where Unicaption fits

Unicaption is a real-time captioning and translation copilot built for interpreters, which is why its privacy design starts from what you'd want to be able to tell a client.

- ✓ **End-to-end encrypted sessions**
- ✓ **Audio is never stored**
- ✓ **Transcripts are deleted when the session ends**
- ✓ **Session content is never used to train models**
- ✓ **Nothing joins the call:** it listens through your microphone or system and tab audio, so there's no bot in the participant list
- ✓ **Live captions and side-by-side translation** in 60+ languages, in the browser, on macOS and Windows, or with the Chrome extension

That's the extent of the privacy claim, and it's the one to cite. Unicaption doesn't hold any compliance certification, so for regulated settings, follow your client's own rules about which tools are permitted. For related reading, see [HIPAA-compliant live captioning](/blog/hipaa-compliant-live-captioning.html) and [live captioning and GDPR in the EU](/blog/gdpr-live-captioning-eu.html).

## Two-minute pre-job checklist

- I know where the tool sends data and what it keeps.
- I've confirmed my content isn't used for model training.
- My client and agency permit this tool on this job.
- I've disclosed it where disclosure is expected.
- I'm not pasting confidential documents into a general chatbot.
- I know what I'll do if the client says no: work without it.

## Frequently asked

**Can interpreters use AI tools without breaking confidentiality?**
Yes, if they check how the tool handles data and the client and agency permit it. Confirm where audio and text go, whether anything is stored or used for training, and who can access it. Unicaption is end-to-end encrypted, never stores audio, deletes transcripts when the session ends, and never uses content to train models.

**Is it okay to paste assignment documents into ChatGPT or Claude?**
Only when your contract, the client, and the tool's current data terms all allow it. Confidential contracts, case files, and patient information generally shouldn't go into a general-purpose chatbot. Build glossaries from public information or your own notes when you can't confirm the terms.

**Do I have to tell the client I'm using a captioning tool?**
Your agency contract and the client's rules decide, but disclosing a tool before the job is good practice and avoids disputes. Describe it accurately: a live reference that shows text on your screen is different from a tool that records or summarizes the conversation. If the client declines, work without it.

**Does Unicaption record or store the conversations I caption?**
No. Unicaption never stores audio, deletes transcripts when the session ends, encrypts sessions end to end, and never uses session content to train models. It runs on your own device through the microphone or system audio, so nothing joins the call.

---

*[Unicaption](https://unicaption.com) is a real-time AI captioning and translation copilot for human interpreters — live captions and translation in 60+ languages, end-to-end encrypted, nothing stored. 30 free minutes every week, no credit card required.*