Recording & Consent Notice

Why this page exists

SubLive processes other people's speech. In most places that is regulated, and the rules are not the same everywhere. This page explains the landscape in plain language so you can decide what you need to do before you press start.

This is guidance, not legal advice.

What actually happens technically

SubLive reads the audio your browser is already playing on the tab you choose. It does not join the call as a participant, and the other people on the call receive no automatic notification from us.

Your microphone is used in one case and only if you ask for it: when you have switched on saving a conversation and then separately switch on adding your own replies to it. That is a second, separate choice, your browser asks you for microphone access before it takes effect, and you can switch it off at any time. Your own speech is never translated and never shown as subtitles — it goes only into the transcript you asked us to keep.

Bear in mind what a microphone hears: not only you, but anyone speaking in the room around you, including people who are not on the call at all. Switching this on is a decision about them too.

Audio is processed in memory and never stored — your own included. In the shared tier it is sent to third-party AI providers for processing; in the local tier it is not.

One situation, and it is the harder one

Subtitles on a tab (one person). You choose a tab and we transcribe and translate whatever plays on it — including the voices of people in a call who never installed anything and were never asked. We physically cannot ask them: they are not our users and we have no way to reach them. This is why the notice you accept before starting says plainly that making sure you are allowed to do this is your responsibility.

There is no arrangement here where we obtain that consent for you. A shared session that every participant joins and accepts for themselves would be the cleaner situation by a wide margin, and it is not what SubLive does: every session is one person capturing one tab.

We record which notice you accepted, when, in which language it was shown to you, and under which version of this text.

United States: one-party vs all-party consent

Federal law and most states permit recording a conversation if one party consents — that can be you.

A number of states require the consent of ALL parties. These include California, Florida, Illinois, Maryland, Massachusetts, Michigan, Montana, Nevada, New Hampshire, Pennsylvania and Washington. The list changes; verify before relying on it.

Where any participant is in an all-party state, assume you need everyone's consent.

European Union / EEA / UK

Processing someone's speech is processing personal data. You need a lawful basis and you must be transparent about it.

In practice: tell participants at the start of the call that speech is being transcribed and translated by AI, and be prepared to say what happens to it.

Workplace contexts (recruitment, performance conversations) attract additional scrutiny, and some member states have specific rules on monitoring employees.

What we do to help

• The notice above is a blocking screen in the extension, not a settings checkbox: subtitles cannot start until you accept it, and the server refuses to open a session for an account that has not.
• Every consent is logged with its timestamp, the policy version, IP address and user agent, so you can evidence it later.
• We log the consent, not the audio.
• Transcripts are off by default. Saving one is a separate switch, and turning it on or off is written to the same consent log, with the date.

What we cannot do

We cannot obtain consent from someone who is not our user. In a call, that is usually everyone except you.

We cannot tell you whether your specific situation is lawful. That depends on where each participant is, what the conversation is about, and in an employment context, on rules that have nothing to do with us.

We cannot detect that a conversation is a job interview, a medical appointment or a legal consultation and refuse to run. The product does not know what it is hearing, and building something that did would mean analysing the content of your conversations — the opposite of what this page is about.