Privacy

Surface Scholars (surfacescholars.com) is run by Gateway Futures Inc., 399 Fremont Street, San Francisco, CA 94105. People record and label speech here, and we license the approved data to companies that train and test AI. This page says what we collect, why, who handles it, who receives it, and how long we keep it. You must be 18 or over to use Surface.

• Account and contact. Your email, name, profile, the documents you upload, and, if you add them, a phone number and contact preferences.
• Recordings and work. Audio of meetings, interviews and voice tasks, their transcripts, and the annotations, reviews and comments you write.
• Consent records. What you agreed to and when. For a meeting, also the IP address and browser you joined from.
• Voiceprints. Only if you opt in to voice matching: a numeric fingerprint of your voice, used to catch duplicate or impersonated recordings. Our biometric policy says how we keep and destroy them.
• Details about your voice. If you give them: age band, gender, accent, languages, the kind of microphone and where you record. Buyers see them only if you tick the box to share them.
• Device details. With each recording: the kind of device, browser and system, and the audio settings the browser used. Never the microphone's name or a device id.
• ID check. On projects that require it: the result of a Stripe Identity check, whether you are 18 or over, and a one-way hash of your document so one ID cannot open many accounts. Never the images, the document number or your birth date.
• Payments. The work you were paid for and the payouts sent. Bank and tax details are held by Stripe, not by us.
• Access logs. Sign-ins, downloads and media access, with your IP address cut to its network (the last part removed) and your browser name.
• Work pace and automation checks. For paid annotation and review: how long you took compared with the audio, how much of it you played, and whether a submission matches the machine transcript unchanged. If we turn them on, also whether your browser reports being automated, whether text was pasted rather than typed (counts only, never what you typed), and whether several accounts sign in from the same network and browser. These are shown to the people who manage the project and to staff approving payments.
• Work pattern and integrity checks. When you claimed, saved and submitted work, used to suggest a break after a long stretch and to show the people who manage the project long sessions, late-night work (in your time zone, if you set one) and accuracy dropping within a session. How your saved text changed between autosaves, to spot a large block of words arriving at once, and whether a transcript reads like a speech engine's or a chatbot's output. For staff only, whether your work speed or check results changed suddenly and, if you opted in to voice matching, whether a recording's voice matches your others. If we turn them on, also the typical time between your key presses (one number per submission, never the keys) and whether your account is active from two networks or browsers at the same time.

To run the service, build and license datasets, pay you, keep work honest (fraud, duplicate accounts, automated work), keep accounts secure, and meet legal duties.

If you are in the EU or UK, our legal reasons are: your consent for recordings, voiceprints and voice details; our contract with you for paying you; our legitimate interest in stopping fraud and keeping the service secure; and the law, for tax and payment records.

Buyers: companies that license approved datasets to train and test AI. They get the audio, transcripts and labels, a code for each speaker instead of a name, each recording's device details, and your voice details only if you chose to share them. Never your email, payment details, ID check or voiceprint. Every buyer agrees not to impersonate or identify a speaker, and to delete a speaker's recordings when they are erased. A buyer may have a dataset sent to its own storage or its own Hugging Face account. Each recording carries the uses its project's consent statement allows.

We also share data when the law requires it.

Sign-in cookies only, no advertising or analytics cookies. A session lasts up to 30 days for contributors, and up to 7 days (12 hours idle) for staff, managers, organizations and buyers. A meeting pass lasts 6 hours. Your browser also keeps unsaved drafts and settings like playback speed on your device.

• Vercel (hosting) and Neon (database)
• Cloudflare R2 (file and recording storage)
• Modal (speech processing)
• OpenAI and Anthropic (transcription, summaries, interview and assistant features)
• ElevenLabs (speech to text and the interviewer's voice)
• LiveKit (live calls)
• Stripe (payouts and ID checks)
• Resend (email), Twilio (text message codes)
• Google and LinkedIn (only if you connect your calendar or LinkedIn)

They process data only to provide their service to us. Data is processed in the United States.

• Withdrawn or rejected recordings: deleted 30 days after withdrawal, unless a project sets another period.
• Voiceprints: deleted 30 days after the project closes (a project can set another period), and never kept more than 3 years after your last recording; 365 days if you live in Texas (the state you gave us, or your profile location if you gave none).
• Sign-in codes: 5 minutes. Rate limit counters: about a day.
• Payment records: as long as tax and payment law requires.
• Access logs and records that something was deleted: kept, to prove who saw what and that deletions happened.
• Database backups: gone within 30 days.
• Everything else: while your account is active, or until you ask us to delete it. A legal hold can pause deletion.

You can withdraw any recording, untick voice matching, and edit or stop sharing your voice details at any time. Contributors can use Delete my data under Account, Privacy. Staff confirm the request, then your audio, transcripts, voiceprints, ID check and memberships are deleted, and your email is replaced by a random id in the records we must keep. Data already delivered in a released dataset cannot be recalled; later releases leave it out.

For a copy of your data, to correct it, to object, or anything else, write to rishi@withgateway.ai, or Gateway Futures Inc., 399 Fremont Street, San Francisco, CA 94105. In the EU or UK you can also complain to your data protection authority.

When this page changes, we post the new version here with a new date.