Approach

Protection without surveillance.

It would be easier to build this by watching everything a child does. We don't. A tool that sees everything can be turned against the child it is meant to protect, and every choice below follows from that.

Predator Hunters has protected children and trained parents since 2017. The lab is the newest part of that work, and it keeps the same hard rules. We treat a child's privacy as part of their safety, not something to trade away for it.

So the models are small enough to run on the phone, the system is built to keep nothing, and a person is always the one who acts. These are real constraints, and we hold them even when they make the work harder.

The line we hold

What our research will never touch.

The screen

We never record or stream what a child sees or types. The screen is read on the phone, in the moment, for one job, and then it is gone.

Their location

No GPS, no movement history, no record of who they are with. None of it is collected.

Raw content

No message, image or video is stored or sent. A parent gets a short, redacted signal that something needs them, nothing more.

Anyone else

The tools run on one child's own phone, with consent. Never on the people around them, never on adults, and never in secret.

Principles

Six commitments, in plain words.

On device by default
The models run on the child's own phone. You can route traffic through a filtering VPN, but only to our servers or an instance you host yourself, never a third party, and it never carries raw content.
We remember nothing
The system is built so there is nothing to leak. No raw messages or images are ever kept, and alerts carry no content.
A person always decides
The models raise a concern and a person acts on it. A parent decides for their child, and our editors decide what we report. Nothing is ever an automated accusation.
Block and report, never store
Illegal child-abuse material is blocked on sight and reported to the right authority, as the law requires. It is never stored, served, or made.
Independent and post-conviction
Our journalism reports only on cases that have been to court and are on the public record. We never name anyone before a trial, and we are independent of any police force.
Open methods, closed data
We explain how the models work. We never publish anything that could harm a child, like their data, a grooming dataset, or live weights.
Questions

The questions people ask first.

Is this spyware?
No. It is openly installed, visible on the device, and can be switched off. It only ever runs on a child's own phone, with consent, and never on adults or anyone who has not agreed to it.
What data leaves the phone?
By default, none of the content. The models judge things on the device. When something needs a parent they get a short, redacted alert with no message contents in it. If you turn on the network filter, traffic goes to our servers or one you host yourself, never a third party, and still nothing raw is kept.
Can it read my child's private messages?
It reads what is already on the screen, on the device, so it can catch grooming inside encrypted apps. It never logs keystrokes or passwords, and that text never leaves the phone or gets stored.
Do you keep or sell any of it?
No. We store no raw messages, images or video, and we have nothing to sell. Illegal material is blocked and reported as the law requires, never kept.
Who is it meant for?
A guardian setting it up on a device they own, for a child in their care. It is not a tool for monitoring adults or anyone who has not consented.

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.