Effective 8 November 2026 · Last updated 8 November 2026 · Applies to: Sight, in the United States · This page accompanies our privacy policy and uses the same definitions.

AI and data disclosures

With your household's agreement, Sight has AI models read your camera clips and write them up in plain language. This page sets out, item by item, how that works and what you decide. Where the full explanation already lives in our privacy policy, we link straight to it rather than say it twice — one promise, in one place.

Whether your data trains AI models

Sight does not train AI models. We build no model of our own, and we do not use your clips, stills, descriptions, or face signatures to train or fine-tune any model — ours or anyone else's. Your home is not a training set.

What does happen is this. With your household's agreement, Sight sends each event's clip to another company's AI model — Google's, or for a home that has visitor recognition on and keeps pictures, Pulses' — which reads it and returns the description you see on your timeline. To write your daily brief and answer Ask, it sends text to Anthropic's model. Sight sends them this to get that work done, and for nothing else. Each company, and exactly what it receives, is named in Who else touches your data. What those companies do with it on their own systems is governed by their terms with Sight, and we do not make promises on their behalf here.

If your household has turned on visitor recognition, Pulses keeps the face signatures it makes for your home, so it can tell you that Tuesday's visitor came back on Friday. That set is kept separately for each home and compared only with faces seen at that home, and it carries no name, no email, and no account details.

Each signature lasts for the window your home has set for keeping pictures. Sight checks your home's set at Pulses every hour and asks Pulses to delete any signature Sight no longer keeps. It asks Pulses to erase the whole set when the feature is turned off, when Keep pictures is set to Off, and when your home's data is deleted, and deleting one person's face data asks Pulses to delete theirs.

What choosing not to share affects

Because nothing is used for training, there is no training to opt out of, and no part of Sight behaves differently depending on such a choice. No feature is withheld, and nothing degrades.

Two choices do change what Sight does. The first is AI processing. Say “Not now”, or withdraw later, and Sight stops sending your home's clips and notes to Google, Pulses and Anthropic. Sight keeps working: each event still gets Sight's own short note, marked “Not described — AI processing is off for this home”, your brief uses Sight's own simpler wording, and Ask answers from Sight's own search. A home's events and brief follow its owners' answers; your Ask questions follow yours.

The second is visitor recognition, an optional feature that is off unless an owner of your home turns it on:

• Leave it off — the default — and no face signature is made, and Sight cannot tell you that someone has been before. Everything else works normally: you still get your descriptions, your briefs and your package alerts.
• With it on, in a home that keeps pictures, Sight groups a returning visitor's visits together, named or not, so you can see them as one person and give them a name if you choose.
• Turn it off later and every face signature Sight holds is deleted, and Sight asks Pulses to delete its copy. Your account, your timeline and your notes are untouched — you do not have to delete anything else to withdraw this one permission.

Separately, choosing to keep no pictures means no still and no face signature is written down at all. Sight still reads each event in the moment — that is how the description gets written — but nothing from your camera is stored afterwards. Video is never stored, whatever you choose.

Who can see your moments

No one at Sight watches your cameras. Your clips are read by software, and Sight keeps no video. Nobody on our team opens a customer's video or browses a customer's timeline, and there is no support screen or admin page anywhere in Sight that plays a customer's camera back to us.

So the list of people who can see your home's moments and pictures is short:

• You.
• Anyone you invite into your circle. Household members see the same timeline and the same pictures you do. Choose your circle the way you would choose who gets a key.

Three limits we would rather state than let you assume away. Sight's systems can decrypt your pictures — they have to, because reading the picture is how Sight knows a parcel arrived; this is not end-to-end encryption and we will not call it that. If something is badly broken, a named engineer may need emergency access to production data; that takes two-person approval, it is logged, and we tell the affected home. And because we hold the keys, valid legal process can compel us to produce what we hold; we require valid process, narrow it where we can, and tell you unless we are legally barred from doing so. All three are set out in full, with everything else our promise does not cover — including what happens inside the AI companies' own systems — in the limits of this promise.

The controls you have

These live in the app under Settings → Account & data. Some are for a home's owners, and say so:

• AI processing — read exactly what Google, Pulses and Anthropic each receive, then agree, say “Not now”, or withdraw at any time.
• Keep pictures (owners) — Off, 7 days, 30 days or 365 days. Off is the default, and it means nothing is written down rather than written and swept up later. Shortening the window deletes what is already older straight away, without waiting for a clock; choosing Off deletes every stored picture.
• People Sight recognises — the people your household has named, with the date each was named. Delete any one of them, which deletes their face data at Sight and asks Pulses to delete theirs.
• Turn off face recognition (owners; shown while it is on) — deletes every face signature Sight holds and asks Pulses to delete its copy. Your account stays.
• Export your story (owners) — a portable copy of what Sight holds, described under Seeing and reviewing what Sight holds.
• Delete account — anyone can delete their own account. A home you are the only owner of goes with it.
• Delete all my data (owners) — destroys the home's encryption key first, which makes its stored pictures unreadable immediately, then removes everything the home owns, asks Pulses to erase the home's face signatures, and reports back what was removed.

Elsewhere in the app, under Settings → Camera & connection, each camera can be switched off individually, so Sight stops processing it while your other cameras carry on.

Outside the app: you can unlink Sight from Ring at any time by removing Sight in the Ring app. The moment you do, Sight deletes the Ring sign-in it holds and can no longer reach your cameras. Unlinking at Ring and deleting at Sight are two different actions, and a complete removal means doing both — see Deleting your data & unlinking Ring.

Seeing and reviewing what Sight holds

Your timeline is the record. Every description Sight has written for your home is there to read, with the time and the camera it came from, and each one can be opened and checked against its picture — from your home's own store if you keep pictures, or fetched from Ring for that one view if you don't.

Export your story gives a home's owner a portable copy on request, and anyone else can ask us for theirs. It contains your notes, briefs, rhythms, settings and account records, plus a line for every stored picture saying when it was taken and when it expires. It does not contain the picture files themselves, and it does not contain face signatures — a signature is a lock-in of someone's face, and we will not hand one out over email. We would rather say that plainly than let “export everything” imply more than it delivers.

People Sight recognises shows the other half: who your household has named, when that happened, and how many times Sight has seen them.

Every moment also carries Sight's confidence in words rather than a score you would have to interpret, and where Sight could not describe something — the video was no longer available, AI processing is off for your home, or there was nothing it could read — it says so instead of presenting a guess as an answer.

When Sight's AI changes

The descriptions on your timeline are produced by named, pinned model versions, and the model that produced a given description is recorded alongside it. Models do not change underneath you as a side effect of something else; changing one is a deliberate act.

When we change what Sight's AI does in a way that affects you — a new detection capability, a change in what gets described, a change in how accurate something is, or a change of provider — we record it on this page and update the date at the top, and where the change is significant we tell you directly. A change in which companies receive your data, or in what they receive, also changes the AI-processing screen in the app, and anyone who agreed to the earlier wording is shown the new one. This is the same commitment we make for changes to the privacy policy, and for the same reason: a change you only discover by noticing your app behaving differently is not a change you were told about.

Where a new capability involves collecting something we were not collecting before, it does not arrive switched on. Visitor recognition is the example: it needs a separate consent screen, accepted by an owner of the home, before it can do anything at all, and it stays off until that happens.

Recorded changes. 8 November 2026 — Sight stopped keeping video: each clip is now read and then discarded. Pulses is now named as the company that reads clips instead of Google for homes that have visitor recognition on and keep pictures. The app now asks for your household's agreement to AI processing, and a “Not now” or a withdrawal stops it.

Questions

Questions about anything on this page, or want to exercise a privacy right? Reach us at privacy@getsight.app, or through support.