Skip to content

Ethics & Privacy

Mission

Colophon is a voluntary writing support tool: it helps students and writers document their process and declare their AI use transparently. It is neither a surveillance system nor an AI detector.

The goal: give students a way to prove their effort, and teachers a way to see the process.

AI-text detectors guess whether a student used AI, and do so unreliably, with documented bias11 AI-text detectors have well-documented false-positive problems, including disproportionately flagging non-native English writers. Colophon avoids this by never trying to detect AI use after the fact. It logs the process as it happens instead.. Colophon takes a different approach: it records the writing process itself, every edit, paste, and AI interaction, in a tamper-evident log that can be shared voluntarily to demonstrate how a document was created. It works like a mirror, showing you your own process.

Two layers, not one

The core tool is for the student doing the writing: it shows your own writing process back to you and your teacher. The research layer is separate and optional: only if your class agrees, an anonymized version of that same process data can also inform a published research report on how people write with AI. Using Colophon never implies agreeing to the research layer; see Consent for exactly how that choice works.

Guiding principles

Adapted from Nine Principles for the Use of AI in Education, published by the UNESCO Netherlands National Commission (2025)22 Jon Verriet & Marieke Brugman, Negen principes voor het gebruik van AI in het onderwijs, UNESCO Netherlands National Commission, 2025. A national-commission adaptation, not a global UNESCO Paris publication, so it's cited as such rather than as "UNESCO's" principles generally., alongside UNESCO's global Guidance for Generative AI in Education and Research (2023).

Principle How Colophon applies it
Right to privacy No personal data leaves your device unless you explicitly export and submit it; see Privacy architecture.
Non-discrimination Colophon does not judge or classify writing; it records activity, not a verdict on the writer.
Transparency & explainability Every event type is documented in plain language on what gets recorded; nothing is a black box.
Human oversight You control the session: start, stop, export, and (for the research layer) withdraw at any time.
Accountability You own your own data. Colophon records process, not a judgment about the content you wrote.

On trusting fluent AI output

"On the Dangers of Stochastic Parrots" (Bender, Gebru, McMillan-Major & Shmitchell, 2021)33 Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? FAccT '21. argues that large language models produce fluent-sounding text without any grounding in meaning or intent. The real risk is not the model; it is people over-trusting fluent output as if it were understanding. Colophon's design leans the other way. Instead of assuming the human/AI boundary away, it makes that boundary visible and inspectable, for the writer and for anyone they choose to show it to.

Designed with, not certified against

Colophon is designed with the principles of the EU's GDPR, the US's COPPA (in its school-context provisions) and FERPA, and South Africa's POPIA in mind: local-first storage, data minimization, purpose limitation, and a real right to access, correct, and delete your data. This is a design commitment, not a legal compliance certification. Like our privacy policy and terms, it hasn't been reviewed by outside counsel yet, and we'd rather say that plainly than overclaim.

Sources

Ethics & Privacy Privacy architecture