Your attention.
Your information.

Your name or nickname is required. For general interest, email, phone number, interests and comments are optional. A Lab request requires a verified email account and permission to reply about that request and its invitation. Phone and comments remain optional. We record your contact choices, a submission reference and the submission time and, for a Lab request, the account ID, verification time and when you gave permission to reply. If you submit while signed out, this browser remembers the reference so it doesn’t ask you again. Clearing browser data removes that reminder. The reference does not give access to your details. Please do not include sensitive personal information or recordings.

We use submissions to understand interest, plan early testing and consider offers to help. Providing contact details does not subscribe you. Email preferences offer three choices: no update emails, all four topics (product, testing, Lab news and ways to help), or only the topics you select. A verified email and an explicit save are required. Existing product-only subscriptions remain product-only. Permissions given on earlier forms do not add anyone to email groups. Choosing No update emails stops update subscriptions, including general invitations to future opportunities. It does not stop replies to a specific Lab request you have asked us to handle, and does not remove Lab access. Phone contact still requires its own permission. Necessary account and security emails are separate. Submitting a Lab request never adds, removes or changes an update subscription. Research participation and recording consent are separate from applying.

If you apply to the Lab, create an account, request updates or manage preferences, Supabase maintains an authentication record for your email. You can use a website password or an email sign-in code or link. Passwords are handled by Supabase Auth, not stored in our interest submissions. Your sign-in tokens are stored in this browser’s local storage so sign-in survives closing and reopening it. While you’re signed in, this browser also remembers whether the account has beta access or a request waiting, so the beta buttons show the right words at once; signing out or deleting the account clears it. The website signs this browser out after 24 hours without a new page load. Loading or refreshing a page before expiry starts a new 24-hour window, and a Lab recording or sound check running in this browser renews it, so the recording isn’t cut off. This applies to this browser and website address; private browsing or clearing browser data may end your session earlier. Sign out when using a shared device. Verification emails are sent through our GoDaddy/Titan mailbox. A verified email or account does not, on its own, grant Lab or administrator access. You can unsubscribe at email preferences.

Every registered account has one shared profile across Protecting Minds. Display name, preferred language and timezone are optional. We do not fill these from old guest forms. Account IDs link your own signed-in submissions, update preferences and Lab membership; matching email addresses alone do not prove ownership of a guest submission. Profiles do not grant permissions or subscribe you to updates. You can edit or clear optional profile details in My account.

We keep a record of changes to saved update choices and Lab request-reply permission, including account references, timestamps and the applicable wording version. The first entries, made when this history began, show what was already stored, not a new permission or a reconstructed history. Authorised administrators can manage explicit update-email blocks; closing a form is separate from blocking its address. Blocks made earlier, when closing a form also blocked its address, still apply.

Website submissions are stored in Supabase, currently in its Seoul region, separately from Lab research records. Our website host, Supabase and email provider process information needed to operate these services, including technical security logs. Only authorised administrators may review submissions. We do not sell this information or use this form for advertising tracking.

So that we notice new submissions, once a day, at 09:00 India time, our server emails a summary of the forms submitted the day before to our own mailbox, protectus@protectingminds.com, and to no one else. It lists each interest form and Lab request as you sent it (your name, email and phone if you gave them, what you’re interested in, your comments, your contact choices and the time), the email-update choices saved that day (the address and its topics), and each beta request (the account’s email, how many times it asked and the request’s status). It holds no passwords or recordings. Of the Lab it gives only the day’s totals (recordings and their hours, sections contributed, AI analyses and their cost), never a member’s name or email. It is sent every day, also on a day with no submissions.

The summary goes from Supabase to our GoDaddy/Titan mailbox over an encrypted connection and stays there as our working copy. Deleting your account doesn’t delete these emails. Our review of interest records and your removal requests cover them too: write to us to have yours removed.

When you send a Lab request or ask to join the beta, our server also emails our own mailbox, protectus@protectingminds.com, as it arrives, so that we answer sooner. For a Lab request it holds what you sent (your name, the account’s email, your phone if you gave it, what you’d like to help with, your comments, whether you allowed a reply, and the time); for a beta request, the account’s email, how many times it asked and the time. Its link opens the request in our private Observatory and carries a reference, not your details.

When we decide, we email you the answer from that mailbox: always for a beta request (approved, declined, or your access ended), at your account’s verified email; for a Lab request only if you allowed a reply about it (an approval brings your invitation code). Nothing else, no tracking, and never to an address with an email block, an unverified address or a deleted account. We record whether each email went, with references, not addresses. Deleting your account deletes that record. It doesn’t delete these emails in our mailbox: write to us to have yours removed.

We review interest records after six months and remove information no longer needed for early testing or collaboration. This review is manual, not an automatic deletion guarantee. You may request removal or withdraw contact permission at any time. For a name-only submission, include the reference shown after submission. Choosing No update emails stops update subscriptions but does not by itself delete your interest, Lab request or authentication record. You can withdraw request-reply permission from your saved Lab request on the Lab page, or contact us for help. This does not change update preferences or Lab membership. Provider backups and security logs may retain information for their applicable retention periods.

Any signed-in account can ask to join the beta of our consumer apps on the web app’s page: the web app first, the phone and TV apps later. Asking needs a tick for the beta terms: you’re 18 or older, it’s a beta version that may get things wrong, the sound always comes back with Restore or the TV remote, and access can be removed. When the terms change, testers tick the new ones once on their next visit. The tick is not consent to research and contributes nothing to it.

We store that your account asked, when and how many times, the version of the terms and when you ticked them, when you withdrew a request or left the beta, and whether an administrator approved, declined or removed your access, when, and which administrator. A history of these changes keeps account references, not email addresses. Both are stored in Supabase, currently in its Seoul region. You see only your own status; administrators, signed in with an authenticator app, see each request with the account’s email, and our mailbox gets an email about each new one with it as it arrives (see Emails about requests). We email you the answer, and it shows on the web app’s page.

The web app analyses the sound in your browser as it plays: nothing is recorded or sent, in the beta or out of it. If you’re a tester, your browser downloads the detector an administrator has published for testers, a model file with no recordings or labels, and keeps a copy. The copy is deleted when you leave the beta or sign out in that browser. If your access ends or no detector is published any more, it is deleted the next time the web app checks in that browser. The web app’s log of what happened (times, scores and your marks, never sound) stays in that browser tab until you close it, and is never sent.

You can withdraw your request, or leave the beta, on the web app’s page at any time. After a decline or a removal you can ask again after 7 days. Deleting your account deletes your beta record and its history. It doesn’t delete the emails about it already in our mailbox.

We’re building apps for Android phones, iPhone and Android TV. The Android phone and TV apps are private test builds for a few testers, and the iPhone app is still being built; none is in any app store. This section describes them as they are built today; it will change before they are released.

The phone app listens only when you start it, or start a sound check, and only while it is open on the screen. It listens to the TV’s own sound through the TV app, which you approve on the TV for each session, or to the phone’s microphone if you choose it; it never changes between the two by itself. It compares the sound with a library of known ads on the phone, using Apple’s ShazamKit. The sound is kept in the phone’s memory only for that: it isn’t saved, and the app doesn’t send it to us. Optional cricket cues use the phone’s own on-device speech recognition; the words aren’t saved or sent, and they never mute on their own. The app mutes the TV only after matching the same known ad twice in a row, at most once each time you start it, and the sound comes back only when you tap Restore, or with the TV’s remote.

The TV app runs only when you start it on the TV. It mutes the TV when your paired phone, or the web app, asks. If you approve it on the TV (Android 10 or later), it also shares the TV’s own sound with the paired phone for the time you choose, up to 8 hours, with a notification and Stop on the TV. For that, Android asks you on the TV for the app’s permission to record audio and for approval to capture the screen. The app still never opens a microphone or records video, doesn’t save the sound, and can’t capture sound that the playing app or the TV protects.

The phone and the TV talk directly over your home network, after you pair them with a code the TV shows. The TV app answers only devices on that network. Messages don’t go through the internet or our servers, and each one is encrypted and checked with the pairing code (AES-GCM). The link carries the mute and Restore commands and, when you approve it, the TV’s sound. The code stays on both devices; making a new one on the TV ends earlier pairings. This is a test design that hasn’t been independently audited. The web app and the browser Lab can use the TV app too, through a small connector you run on your computer.

The apps need no account. Each phone app keeps a history of its troubleshooting records on the phone, and you can share or save a report from it yourself. In either phone app you can also sign in and turn on diagnostic sharing, which sends those records for new sessions to us (see Optional session diagnostics).

While you fill in a Lab request, we keep a draft in this tab’s session storage for up to one hour after the last edit. It can include your name, email, optional phone and message. An expired draft is removed while the page is open, or the next time the page wakes or reloads. It is not separately encrypted and does not travel to another device. In this tab it is restored when you are signed out, or signed in with the same email as the draft; a draft for a different email is not loaded. Clearing this tab’s browser data removes it. A restored draft is never submitted automatically. A confirmed submission clears the draft in that tab.

Expressing interest or signing in does not grant Lab access or consent to research. Approved adult participants may open the Lab using their website account or a Lab app. Active membership is checked before protected actions and periodically during use. The Lab records only after you choose a source, give permission and start a session. If you schedule a recording, it starts automatically at the time you set, while the page stays open. Until then the Lab keeps the audio source you chose open and only measures its sound level: nothing is saved, uploaded or sent for AI analysis. The Lab does not mute a television.

Recordings and session notes stay on your device, and in your private Lab cloud space when My Lab cloud is on for a recording (see below), until you choose to contribute sections to the research dataset, ask for AI labels for sections you pick, or turn on automatic OpenAI labelling for a new recording. Each of these needs its own consent. An audio file you import into the Lab is treated the same way: it is cut into parts on your device, and none of it leaves your device unless you keep My Lab cloud on, ask for AI labels or contribute it. Local drafts are not separately encrypted from the browser profile. Signing out hides them but does not erase them; clearing app or browser data may remove them. Keep private conversations out of recordings and obtain the necessary broadcast and voice permissions.

A contribution includes the audio and its labels: your own labels, or AI labels you chose to contribute without checking them. Labels made with AI help stay marked as model-assisted, and we record whether the labels came from you or from AI whenever that record can be saved. A contribution also includes your account reference, consent acknowledgement, session context such as the session name, channel and language you entered, and listening-quality information. It is stored privately in Supabase for authorised research review and advertisement-detection model development. Only authorised research members may access contributed data under their assigned roles. Recording alone does not contribute audio to the research dataset or train a model.

Contributed clips are kept for the research dataset until you delete them, your account is deleted, or we remove them, for example after a review or at your request; they are not deleted after a fixed period. A clip rejected in review, or whose audio check fails, is kept out of training and exports and is never deleted automatically: it stays until you or an administrator delete it. You can delete contributed clips through the Lab or contact us for help. Deletion cannot recall a previously distributed research export or undo a trained model. Detectors trained from contributions are kept for Lab administrators only, with their test reports, which name the recordings they were tested on. A daily cleanup finishes the deletions you ask for, and removes uploaded audio that no contribution or recording claims once it has been left over for 14 days. Each day it deletes at most 500 files you asked to delete and at most 50 others, so finishing a large deletion can take several days. If more than 50 such left-over files are found at once, none of them is deleted until we have checked them. If a cleanup fails, operators see it; being due for cleanup does not guarantee deletion at an exact time. Provider backups and technical logs may have their own retention periods. Participation is optional and requires separate contribution consent.

When you start a browser Lab recording, My Lab cloud is on unless you switch it off, and the recording consent then says so; you agree to it for each recording. Each 15-second part of the recording is saved to a private storage area of your account as it is recorded, with its time, size and checksum, together with the recording’s name, channel, language, capture-quality events and how it ended (with your reason, if you add one), and the labels you give its parts (including records of AI suggestions shown to you). This lets you review, label and contribute the recording from any browser where you use the Lab. Older recordings are saved there only if you choose Back up to the cloud and agree to the same text.

The cloud space is stored in Supabase, currently in its Seoul region, separately from the research dataset. Only you, signed in with active Lab membership, can open, add to or delete these recordings. They are not shared with the research team, not used for research or model training, and not included in research exports or the review desk; only a section you contribute is added to the research dataset, under its own consent. Administrators see totals only, such as the space used. As with any stored data, the project’s database owners could technically reach it.

Parts wait in your browser while it is offline and upload when the Lab is back online, while a Lab tab is open. Your browser keeps its own copy while it has room; parts the cloud has confirmed may be cleared from the browser to make room, and are fetched again, and checked against their checksums, when you play or use them.

Cloud recordings are kept until you delete them. You can delete a recording from this browser, from the cloud or everywhere, and delete the cloud audio of contributed sections on its own (the research dataset keeps its copy). A recording deleted from the cloud leaves a small record of its reference and deletion time for about a year, so no browser uploads it again unless you choose to back it up again. If you leave the Lab, your cloud recordings are kept for 30 days in case you return. The daily cleanup then deletes them, at most 500 files a day, which takes about a day for every two hours of recording, or longer when other deletions are waiting; you can delete them at once when leaving instead. A Lab administrator’s recordings stay until they delete them. Deleting your account deletes your cloud recordings. Provider backups and technical logs may have their own retention periods.

Optional OpenAI labelling is separate from research contribution and diagnostic sharing. When you ask for AI labels yourself (Get AI suggestion for one section, analysing or retrying several sections, or Analyse again), you consent once for the sections listed, and those whole sections, each up to 60 seconds, are sent through our server to OpenAI. There is no separate preview, and listening first is optional. Retrying a failed request or choosing Analyse again sends the same audio to OpenAI again as a new request; earlier results and their costs are kept. Alternatively, you can enable automatic analysis before a new browser recording and consent to sending successive 15- or 60-second batches as they are saved, without previewing each one. This option is off by default and needs fresh consent for each recording. Both kinds of request wait in a queue kept in this browser and are sent a few at a time while the Lab is open in this browser (also while you are on other pages of this site after opening it, or the tab is hidden), after a page refresh, or from another Lab tab of the same account, until they are sent, expire or you stop them. You can stop automatic sending without stopping the recording; batches not yet sent are then cancelled, but any already on their way finish. When a recording stops, for any reason, its last batch (the audio since the last full batch, if at least a second) is still sent once under the same consent. A batch held back by a limit or a lost connection is sent later as the same request, so it is not sent or charged twice. Where the consent you agree to says so, a batch whose answer can’t be used, or that never reached OpenAI, is sent again automatically as a new paid request, at most 3 more times; otherwise a failed batch is sent again only when you ask. Opening or playing a saved recording does not send audio to OpenAI. Do not send private conversations or content you do not have permission to share. Our server handles the audio in memory and does not add it to research storage. OpenAI processes the submitted audio under its API data policies.

We retain account-linked request and recording references, audio hashes and time ranges, consent version and time, model and prompt versions, validated label suggestions, usage estimates and safe error codes. We also record when suggestions are shown to you or used as labels, so labels made with AI help are not mistaken for independent human labels. Suggestions never become reviewed labels or training data on their own: only sections you choose to contribute are added to the research dataset. You can remove saved results through the Lab or My account, even after clearing local audio or leaving the Lab. Removing results keeps minimal cost and model-exposure records. Deleting your account removes results and account links but keeps anonymous cost records. Stopping a request cannot recall audio already sent; copies held by providers and earlier research exports follow their own rules. Saved results stay until they are removed; the research-audio cleanup does not apply to them.

The pilot apps and browser Lab keep a limited set of troubleshooting records on the device for Start attempts, setup, listening health, control actions and failures. These contain random session and command IDs, build versions, timing, fixed event codes and numeric counts—not recordings, transcripts, programme names, email addresses or pairing credentials. The records are limited to 10 MB and expire after 7 days; expired ones are cleared when the app next opens or writes. Clearing device data may remove it sooner.

Cloud diagnostic sharing is optional and off until you enable it on that device. Only new sessions are shared; earlier local history is not uploaded. Your signed-in account is linked privately on the server to enforce access, consent and quotas. Turning sharing off discards pending diagnostic uploads on that device; it does not recall previously accepted records. Cloud session diagnostics are eligible for deletion after 30 days, with 10 MB per account and a 50 MB shared operational limit. Permission and operator-access audit records are eligible after 90 days. Authorised, separately approved operators must verify with an authenticator before inspecting diagnostics; this permission does not grant research-audio access. Diagnostic reports exclude account identity. Audio contribution and update-email choices are separate.

During the pilot, storage allowances are shared by all Lab members rather than set per account. Contributed cloud audio shares one 20 GB allowance (about 170 hours), including space reserved for uploads in progress. My Lab cloud recordings share a separate 60 GB allowance (about 520 hours); when it is full, new parts stay in your browser and nothing is deleted automatically. In the Lab apps, recordings on the device are limited to 1 GB per installation and need 100 MB of free space. In the browser Lab, recordings are limited to 250 MB for this website, and to the storage your browser has available. If you also save recordings to a folder on your computer (Chrome or Edge on a computer), each clip is written there too and the Lab sets no limit of its own on the browser copy: recording stops when this browser’s storage or the folder runs out of space. The folder copy stays on your computer; nothing is uploaded from it. These are maximum allowances, not guaranteed reserved space. A full allowance blocks new recording or uploads without automatically deleting your existing audio. The app tells you when usage cannot be verified instead of treating it as zero.

Lab apps for Android phones and iPhone were made for the Lab’s first tests and are paused: they aren’t offered for download, and the Lab works in the browser. An Android Lab app that is already installed follows the same rules as the Lab in the browser: its recordings stay on the phone, and contributing, AI labels and diagnostic sharing each need their own consent. The iPhone Lab app was never released. This section will be updated before the Lab apps come back.

Leaving the Lab disables research participation but keeps your shared Protecting Minds sign-in, profile and update preferences. Previously contributed clips are not automatically deleted; contact us for removal if you cannot access them. My Lab cloud recordings are kept for 30 days after you leave, then deleted, which takes about a day for every two hours of recording (a Lab administrator’s stay until they delete them); you can choose to delete them at once when leaving. Returning requires a fresh approved invitation. Deleting your Protecting Minds account is a different, permanent action affecting your shared sign-in and associated data across the website and apps. The confirmation explains which cloud and device data are removed, and what it doesn’t reach, such as the emails about your forms and requests already in our mailbox. Administrators may need to arrange removal of their administrative role first. A failed deletion is not reported as complete.