I am a Senior Research Engineer working on systems theory and digital provenance.
The Problem: The current industry standard for AI disclosure is a binary ("Did you use AI? Yes/No"). This is insufficient for professional workflows. It fails to distinguish between using an LLM for spell-check (Augmentation) vs. using it to generate raw text (Generation).
The Solution: I developed the Authorship Transparency Statement (ATS) Framework v1.0.
It anchors disclosure on a mechanical "Bright Line": The Origin of the First-Pass Prose.
- ATS-0/1: Human writes the draft (AI used for refinement).
- Minimal JSON and Extended JSON-LD schemas for embedding this metadata into digital files.
- An institutional intake template (CSV).
I released v1.0 specifically to stress-test the logic against edge cases. My goal is adoption and utility, not dogmatic adherence to my first draft. If you have suggestions for optimizing the schema or the classification logic, I’d love to see them opened as Issues or Pull Request on the repo.
Happy to answer questions about the "Bright Line" logic or the C2PA interoperability.
djeffbee•2w ago
I am a Senior Research Engineer working on systems theory and digital provenance.
The Problem: The current industry standard for AI disclosure is a binary ("Did you use AI? Yes/No"). This is insufficient for professional workflows. It fails to distinguish between using an LLM for spell-check (Augmentation) vs. using it to generate raw text (Generation).
The Solution: I developed the Authorship Transparency Statement (ATS) Framework v1.0.
It anchors disclosure on a mechanical "Bright Line": The Origin of the First-Pass Prose.
- ATS-0/1: Human writes the draft (AI used for refinement).
- ATS-2+: AI generates the draft (Human edits).
- ATS-4+: Agentic Generation.
The Repo includes:
- The full Technical Standard (https://doi.org/10.5281/zenodo.18091713).
- Minimal JSON and Extended JSON-LD schemas for embedding this metadata into digital files.
- An institutional intake template (CSV).
I released v1.0 specifically to stress-test the logic against edge cases. My goal is adoption and utility, not dogmatic adherence to my first draft. If you have suggestions for optimizing the schema or the classification logic, I’d love to see them opened as Issues or Pull Request on the repo.
Happy to answer questions about the "Bright Line" logic or the C2PA interoperability.