Provenance
About the images.
Dose 3 of our curriculum teaches eleven-year-olds to ask who made this before they believe a picture. It would be a poor look to teach that and then not answer it ourselves.
So every image on this site is labelled. Where you see this mark, a machine made the picture:
AI
The symbol
It is not ours, and that is the point.
The pin is the Content Credentials mark, published by the Coalition for Content Provenance and Authenticity (C2PA) and released as an open, freely adoptable icon. Adobe, Microsoft, Google, LinkedIn and a number of camera manufacturers already use it. A standard only works if people meet the same symbol everywhere, so inventing our own would have been vanity.
In the wider ecosystem, the pin signals that a file carries a cryptographically signed manifest describing how it was made — a “digital nutrition label” naming the tools used, whether generative AI was among them, and what was edited along the way.
What ours does and does not do, precisely. Our images do not yet carry signed C2PA manifests embedded in the files. The mark on this site is a disclosure: we are telling you, on the page, that the image was generated. That is weaker than a signed manifest, because you are trusting us rather than checking mathematics. Embedding real manifests is on our list. Until then we would rather label honestly and say what the label is worth than imply a guarantee we have not earned.
The inventory
Every image, and where it came from.
| Images | Origin |
|---|---|
| The three news photographs on the home page |
Real photographs, not ours. Each is the publisher’s own image for the story it links to, credited on the card — UN News, NPR and the NYC Mayor’s Office. No mark, because no machine made them. |
| The Webspinner Foundation image on the About page |
The Foundation’s own brand artwork, taken from webspinner.foundation. Not ours and not marked, for the same reason the news photographs are not: we did not make it. |
| The AInoculate mark masthead, footer, favicon |
The Foundation’s brand mark. Cut from one source file by tools/make-brand-assets.py; not generated per-page. |
| The classroom and lesson photographs throughout the curriculum, thesis, support and about pages |
Generated with gpt-image-1 and marked. They depict the teaching we describe; no real classroom, school or child is pictured, and no image portrays an identifiable person. |
| The diagrams reach of a ban, Ground Truth, the agathogen, the five doses |
Generated with gpt-image-1 and marked. Every number and label in them was checked against the source before publication — see the note below. |
| The charts book bans, cellphone bans, recall, the guardrail result |
Drawn from the data as code, not generated. They are inline SVG produced by tools/make-figures.py directly from the published figures, so the bar heights are arithmetic rather than an impression of arithmetic. Each carries its source beneath it. |
The rule we work to
A machine may draw. It may not testify.
Generated imagery is fine for depicting a kind of scene — a lesson of the sort we run, a desk of the sort a teacher works at. It is not fine for anything a reader could mistake for evidence.
- No generated image stands in for a real event. The news placards use the publishers’ own photographs precisely because a convincing generated picture of a real event is a forgery, whatever the intent.
- No chart is generated. Charts are drawn from the numbers in code. An image model will render a confident bar of the wrong height, and a wrong number rendered beautifully is worse than no chart at all.
- Labels inside generated diagrams are proof-read against the source. One of ours was regenerated three times because it printed the decimal comma “1,2 M”, then duplicated a card and dropped another. Those are exactly the confident errors Dose 2 teaches children to catch.
- No identifiable person is generated. No image on this site depicts a real child, teacher or school.
- Nothing depicting the harms is generated. Dose 3 covers the deepfaking of children; its illustration shows the method of detection, never the injury.
Sources: C2PA — Introducing the official Content Credentials icon · contentcredentials.org
Why bother labelling at all.
Because we are asking schools to teach children to interrogate the provenance of what they see, and an organisation that makes that argument while quietly publishing unlabelled synthetic imagery has already lost the argument.
It costs us nothing to say which pictures a machine made. If it ever costs us something, that will be the moment it matters most.