Skip to content
Coverage
What we catch, and what we can't.
No tool sees everything, and a child-safety tool that pretends otherwise is its own kind of harm. Here is what works on the device today, what is on-screen only, and what is not technically possible. We would rather you trust this page than a feature list.
| Surface | How we handle it | Where it stands |
|---|---|---|
| Photos the child takes | PH Camera checks every frame on the device before it can be saved. | Working in alpha |
| Images on the open web | Filtered in place on the device, so the page keeps working. | Working in alpha |
| Grooming in chat text | An on-device model reads the text already drawn on screen and flags the pattern. | Working in alpha |
| Unsafe video | Detected as it plays and rewritten in place on the device, blurring or muting only the bad moments. | Alpha, our flagship |
| Encrypted (end-to-end) apps | We never touch the wire. We read what is already on screen, on the device, after the app has decrypted it. | On-screen only |
| Apps that pin their certificates | The optional network filter cannot inspect these. We fall back to the on-device screen read, which is lossier. | Partial, by design |
| The optional network filter | Routes non-pinned traffic through our server or one you host. It helps with some apps, not all. | Partial |
| Audio (calls, voice notes) | Not yet. It is on the research bench, not in a shipping build. | Research |
The one line we will not cross
We do not break encryption, and we never will. When a chat is end-to-end encrypted, the only honest way to help is to read what is already on the child's own screen, on their own device, after the app itself has shown it. That is a deliberate limit, not a gap we are hiding.
Backers · partners · researchers
Help us build the AI that keeps children safer.
We are a small, self-funded team, four years into this. If you fund safety research, want to build with us, or want to put these models to work protecting children, get in touch.