Database
AI Watermark Database Changelog
Editorial history of published records. Git remains the technical history. Dates are research dates, not deploy timestamps.
· Updated · openai
OpenAI provenance records reverified against current Help Center, advancing-content-provenance, OpenAI Verify, and Content Provenance API documentation. Image C2PA/SynthID scope refined to supported ChatGPT, Codex, and API outputs with historical coverage limits.
· Added · openai
Public verification capability record added: web verifier and documented Content Provenance API; AWC integration remains not-integrated. not_detected is not a human-authorship verdict.
· Updated · gemini
Google / Gemini records rechecked against current DeepMind, Gemini Apps Help, Gemini API, and Nano Banana family posts. Image SynthID scope now includes documented API models without treating every historical file as identical.
· Added · gemini
Visible watermark row added as conditional / partial and Gemini Apps-only, with documented regional and account exceptions. Not a universal Nano Banana claim.
· Added · gemini
Nano Banana 2 Lite API model-card C2PA claim added. Not extended to every Nano Banana version.
· Added · claude
Claude research records published: statistical text watermark, hidden Unicode ruled out, and C2PA on supported generated files.
· Corrected
Evidence taxonomy hardened so official source authority is independent of claim status (Phase 4.1). Hidden Unicode is official evidence + ruled out, not an official mechanism.
· Added · openai
OpenAI records added from primary documentation: image C2PA, image SynthID, audio SynthID, and an unknown text-watermark row.
· Added · gemini
Google / Gemini records added: image SynthID (including Nano Banana), Gemini Apps C2PA, text SynthID, video SynthID, and scoped audio SynthID.
· Added · adobe-firefly
Adobe Firefly Content Credentials record added. Firefly SynthID remains unknown rather than asserted absent.