For journalists. Protect the source by making fewer copies.
An interview recording is a list of who talked to you, when, and about what. Shield laws protect you. They do not reliably protect a transcription company in another country that happens to be holding the same file.
Use it forLong interviews, reporting on networks you do not trust, and any source who needs the tape to stop existing.
The argumentA company that never received the file cannot hand it over, cannot lose it in a breach, and cannot be gagged about either.
Not forProtecting a seized device. It is not encryption you control, and it hides nothing about the fact that you met.
Transcription is the least glamorous and most time-consuming part of reporting. It is also where the file multiplies.
01The long interview
Ninety minutes with one person produces roughly fifteen thousand words, of which perhaps four hundred will be quoted. The other fourteen thousand six hundred still have to be readable, because the story is usually in the part you did not expect.
Transcribing it by hand costs most of a day. Uploading it costs a copy.
In the app: record, then transcribe with a large model afterwards. Full-text search across every transcript you have, and timestamps so a quote can be checked against the audio before it runs.
02Reporting where there is no network, or none you trust
Border regions, aircraft, protests where the cell network is deliberately degraded, countries where the local network is not a neutral party. A cloud tool in those conditions either fails or uploads at exactly the wrong moment, over the wrong network.
In the app: keep the phone in airplane mode for the whole trip. Recording, transcription, translation and summarising all work with the radios off, because they were never using them.
03A source who wants the tape to not exist
Sometimes the agreement is that you can use what they said but the recording must go. Honouring that with a cloud tool means trusting a deletion endpoint and a backup policy you cannot inspect.
In the app: Advanced Privacy records once, erases the audio the moment the transcript exists, and clears the session on leaving. You keep the words, the voice is gone, and you can say so accurately.
04Verification and fact-checking
Before publication somebody has to check that the quote matches the recording, in context, with what came before it. That is a search problem across hours of audio, and it is why timestamps matter more than pretty formatting.
In the app: the transcript highlights word by word during playback, so jumping to the moment a phrase was said takes a tap. Export with timestamps and speaker labels for the lawyer's read.
05Foreign language interviews
A fixer translating live is a second interpretation layer and a second person who knows who you met. Having the original transcribed in its own language, then translated, keeps the original available for a translator you choose later.
In the app: fifty-nine transcription languages with automatic detection, more than 1,600 through Omnilingual, and a finished transcript translates into fifty-five languages on the device. The translation is a rough reading tool, not a publishable one.
Your protections do not travel with the file
Source protection is built out of things that attach to you: professional privilege, shield laws where they exist, the willingness to refuse. None of that attaches to a transcription vendor outside the EU. A company holding your interview audio has no duty to your source, no tradition of refusing, and often a legal team whose job is to comply quickly and quietly.
It is also a company that may be in a different jurisdiction, may be acquired, may be breached, and may receive a request it is forbidden from telling you about.
What uploading an interview creates
A copy of the audio, a copy of the transcript, an account tied to your identity, a timestamp, and usually an IP address. Together that is a record that you spoke to this person at this time, which is the exact fact a source needs kept secret, separate from anything they actually said.
The metadata is frequently the more dangerous half. Content can be denied. A timestamped upload is harder to argue with.
What Just Transcribe leaves behind
Nothing outside the device. The model runs on the phone, there is no account, and the app makes no request that says an interview happened. It connects twice ever: once to download a model you pick, once for the App Store purchase.
With Advanced Privacy, the audio is erased as soon as the transcript exists and the session is wiped when you leave the screen.
This protects the source from a vendor. It does not protect them from you, from your device being seized or compelled, from your notes, or from the fact that you met in public. Those are operational problems and this app solves none of them.
Why Just Transcribe
The useful test is what exists after the interview that you did not choose to create.
✓No account, so no identity attached. Nothing links a recording to an email address, a subscription or a login history.
✓No upload, so no request to intercept. There is no network call that reveals an interview happened at all.
✓Nothing for a third party to be served with. A company that never received the file cannot be compelled to produce it or gagged about having done so.
✓Advanced Privacy. Audio erased once the text exists, session cleared when you leave. Keep the words, lose the voice.
✓Works with the radios off, permanently. Offline is the normal operating mode rather than a degraded one.
✓Timestamps, speaker labels and exports. SRT, VTT, PDF and text, which is what a fact-checker and a lawyer actually ask for.
✓One payment. No subscription tied to a card, an employer or a name.
A company that never received the recording cannot hand it over, cannot lose it in a breach, and cannot be ordered to keep quiet about either.
That is the only claim this page makes. It is a narrow one, and it still holds when somebody is served with a demand.
Where Just Transcribe sits in your operational thinking
One column of a much larger spreadsheet. Treat it as the column you can close, not as the spreadsheet.
A rough threat model. If the consequences of exposure are severe, get advice from someone whose job that is.
Vendor subpoena or warrant
Addressed. There is no vendor holding anything, so there is nobody to serve and no gag order to receive.
Vendor breach
Addressed. A company that never received your interview cannot leak it.
Network surveillance
Largely addressed for this step: no upload means no traffic that reveals an interview took place. Everything else your phone does is unchanged.
Device seizure at a border
Not addressed. The files are on the device. Consider what you carry, use Advanced Privacy so audio is already gone, and get real advice about crossings.
Compelled disclosure from you
Not addressed, and no software can be. Fewer copies means fewer places, not immunity.
Your own metadata
Partly. The app creates none, but your calendar, call log, location history and messages all still know who you met.
None of this is security advice. It is an account of what one narrow change does and does not buy.
What to tell a source
I am recording this on my phone. It goes to no company and there is no account attached to it. The phone turns it into text and then erases the audio, so within the hour the recording does not exist any more. What I still have is the text and my notes. Say stop and I stop.
Being able to say that truthfully, and put the phone in airplane mode in front of them, does more for a nervous source than any sentence about encryption.
Setting up before a trip
Four things, done before you leave rather than in a hotel lobby.
Download the model at home, on a network you trust. Once it is on the device, the app never needs a connection again. Take the largest model your phone will carry if accuracy matters more than speed.
Turn Advanced Privacy on for source interviews. Records once, erases the audio as soon as the text exists, clears the session when you leave the screen. Make it the default for anything sensitive.
Decide what leaves the phone, and how. Export is a deliberate act. Decide in advance whether transcripts go to an encrypted drive, an encrypted laptop or a newsroom system, and do it the same way every time.
Practise the offline run once. Airplane mode, record, transcribe, export. Doing it once at your desk means you are not learning it in front of a nervous source.
And the part that is not about this app at all: a strong passcode rather than a four digit one, biometrics off when crossing a border if that is your situation, and a plan for the device that does not depend on any single piece of software behaving well.
What Just Transcribe does not protect you from
Being clear about this matters more here than on any other page on the site.
×It does not protect a seized device. The files are on the phone. Device encryption and your passcode are the protection, and both have limits at a border.
×It is not encryption you control. Files sit in the app's private storage with iOS file protection. That is not the same as a container with a passphrase only you know.
×It does not hide that you met. Your phone's other metadata, your calendar and your location history are untouched by any of this.
×No secure sharing. Getting a transcript to an editor is your problem, through whatever secure channel your newsroom already uses.
×No live transcription. Text arrives after the recording, so it cannot help you follow a fast press conference in real time.
×Not audited. No third party security audit has been published. The verifiable claim is behavioural: it produces a transcript in airplane mode, which means the model ran locally.
If your work involves state-level adversaries, this app is a marginal improvement inside a plan that needs to be much larger. If your work involves ordinary confidential sources, removing the vendor is a real and cheap win.
Straight answers
Is this safe enough for a confidential source?
Safer than uploading, and not a complete answer. It removes the third party and the network trace, which are real risks. It leaves the device risk entirely with you: seizure, compulsion, shoulder surfing, a phone left in a taxi. If the stakes are high, treat this as one improvement in a plan built with someone who does operational security for a living.
Could you be forced to hand over my interviews?
No, because we never receive them. There is no server that touches your content, no account, and no log of what you recorded. A demand served on The Dutch Logic Co. would find nothing to produce. That is not a promise about future behaviour, it is a description of an architecture that has nowhere to keep your files.
Does it work in airplane mode for a whole trip?
Yes, once the speech model is downloaded. Recording, transcription, speaker labelling, summarising and translation all run on the device. That is worth testing at your desk before you rely on it in the field, and it takes about two minutes.
How good is the translation for foreign language interviews?
Good enough to know what you have and not good enough to publish. It translates a finished transcript into fifty-five languages on the device, which is the right tool for deciding which twenty minutes of a two hour interview deserve a professional translator. Quoting from a machine translation is how corrections get written.
What about a long recording, does the phone cope?
Recording keeps running when the screen locks or you switch apps, so a long interview is not halved by a phone call. Transcription happens afterwards and takes time proportional to the length and the model size, so a two hour interview with a large model is something you start and walk away from, ideally on a charger.
Next
Test the claim the way a reporter should: download it, fetch a model, put the phone in airplane mode, and record yourself reading this paragraph. If the transcript appears with the radios off, the model ran on the phone, and you have verified the only thing on this page that you should take on trust from a vendor.
Written by the developer, and not security advice. I make Just Transcribe. If you are working on something where the consequence of exposure is someone's liberty, your threat model needs a security adviser and an operational plan, not an app recommendation from a vendor. This page covers one narrow improvement: removing a third party from the chain.
Written 26 September 2026, updated 3 October 2026 for Just Transcribe 1.36. Numbers and sources on the fact sheet.
