← AI Transparency Labels

AI Transparency Label — Vector Legend

Every label has a compact canonical string that encodes all structured dimensions deterministically. Same label, same vector — across languages, formats, and renderings. This page is the dictionary.

Anatomy

<labelId>/AITL:1/I:<inv>/T:<touched>/M:<models>/R:<review>/S:<sources>

Single-value sections use one letter. Multi-value sections (T, M, S) join values with '+' and are sorted to be canonical. An empty section is encoded as '-'.

I: — Involvement

CodeMeaning
Hhuman only
RAI-assisted research
CAI co-drafted, human-edited
GAI-generated, human-reviewed
UAI-generated, unreviewed

T: — What AI touched

Joined with '+', sorted alphabetically.

CodeMeaning
Rresearch
Fsource verification
Cargument critique
Sstructure
Pprose
Ifigures
Kcode
Ltranslation

M: — Models used

Free-form short codes — model name lowercased, spaces replaced with hyphens. Joined with '+', sorted alphabetically. Examples: claude-opus-4.7, gpt-5, gemini-3.

R: — Human review

CodeMeaning
Nno review
Sspot-check
Ffull review
ESME sign-off

S: — Sources

Joined with '+', sorted alphabetically.

CodeMeaning
Hhuman SME
Ppublic web
Iinternal research
Cclient data
Vvendor data

Worked example

CICS-2026-05-001/AITL:1/I:R/T:C+F+R/M:claude-opus-4.7/R:F/S:H+P

Label CICS-2026-05-001 · AITL version 1 · involvement: AI-assisted research · what AI touched: argument critique + source verification + research · models: Claude Opus 4.7 · review: full · sources: human SME + public web.