AI tools: C2PA, watermarks and labels.

Short answer

Of the 25 AI tools in this table, 11 document adding C2PA Content Credentials to what they generate, among them ChatGPT images, Google’s Gemini, Imagen and Veo, Adobe Firefly, Runway and Synthesia, and 7 document an invisible watermark, such as Google’s SynthID or ElevenLabs’ audio watermark. For 10 tools, including several video and music generators, we found no official statement on either, or could not read one.

Official sources: Google Cloud: Content Credentials · Synthesia: Content Credentials and AI labels

AI tools: what their own documentation says about Content Credentials and watermarks
ChatGPT, Codex and the OpenAI API (images)imageDocumented: addsDocumented: adds6 October 2026Page with quotes
Sora (video; discontinued)videoDocumented: addsNot documented3 November 2026Page with quotes
Adobe Fireflyimage and videoDocumented: addsNot documented6 October 2026Page with quotes
Google Gemini image models, Imagen and Veoimage and videoDocumented: addsDocumented: adds3 November 2026Page with quotes
Microsoft DesignerimageDocumented: addsNot documented3 November 2026Page with quotes
Bing Image CreatorimageDocumented: addsNot documented3 November 2026Page with quotes
Meta AI image generationimageNot documentedDocumented: adds3 November 2026Page with quotes
Stability AI (hosted API)imageDocumented: addsNot documented3 November 2026Page with quotes
FLUX (Black Forest Labs)image and videoDocumented: addsDocumented: adds3 November 2026Page with quotes
MidjourneyimageNot documentedNot documented3 November 2026Page with quotes
Amazon Nova CanvasimageDocumented: addsDocumented: adds3 November 2026Page with quotes
RunwayvideoDocumented: addsDocumented: adds3 November 2026Page with quotes
Grok Imagine (xAI)image and videoNot documentedDocumented: visible watermark3 November 2026Page with quotes
ElevenLabsaudioNot documentedDocumented: adds3 November 2026Page with quotes
Synthesiaavatar videoDocumented: addsNot documented3 November 2026Page with quotes
HeyGenavatar videoNot documentedDocumented: visible watermark3 November 2026HeyGen’s FAQ lists watermark removal as a feature of the paid Creator plan; it does not mention C2PA.
Canva AIimageNot documentedNot documented3 November 2026Canva’s AI Product Terms forbid removing provenance or metadata tags, including C2PA metadata, from AI-generated content; they do not say which tags Canva adds.
SunoaudioNot documentedNot documented3 November 2026Suno’s Terms reserve the right to append a fingerprint, watermark or metadata showing the service tier and forbid removing them to hide an output’s provenance; commercial use depends on a permitted download.
Luma (Dream Machine)videoNot documentedNot documented3 November 2026Luma’s Terms reserve the right to embed watermarks, content credentials or provenance metadata and forbid removing them; commercial use requires a paid subscription allowing it.
PikavideoNot documentedNot documented3 November 2026Pika’s FAQ says downloads on current plans, including Free, carry no Pika watermark, and that Free plans do not include commercial use; it does not mention C2PA.
Higgsfieldimage and videoNot documentedNot documented3 November 2026Higgsfield’s Terms say it may embed watermarks, secure metadata or C2PA-based provenance signals and do not warrant that a marking is applied to every output.
IdeogramimageNot documentedNot documented3 November 2026No official statement on C2PA or watermarks found as of 3 November 2026.
Kling AIvideoNot documentedNot documented3 November 2026No official statement could be read as of 3 November 2026: the pages we tried load their text with JavaScript.
UdioaudioNot documentedNot documented3 November 2026No official statement could be read as of 3 November 2026: the pages we tried load their text with JavaScript.
Leonardo.AiimageNot documentedNot documented3 November 2026Not checked: Leonardo’s terms page refused scripted requests (HTTP 403) on 3 November 2026, and we did not try to get around it.

How entries are chosen

  • An entry exists only if we found the statement on the vendor’s, platform’s or regulator’s own page and quoted it. Entries we could not confirm that way are left out rather than guessed.
  • Every quoted sentence was re-checked against the live page on 6 October 2026 by script. Pages that block scripted fetches were read from an Internet Archive snapshot of the same page, and the entry says so.
  • We report what a source says, nothing more. “Not documented” means we did not find it in the pages we read; it is not a statement that something is absent.
  • On 3 November 2026 the quotes of 11 of the 13 earlier tool pages were checked again against the live pages; OpenAI’s and Adobe’s help pages refused scripted requests (HTTP 403), so those two keep their 6 October 2026 readings, shown with that date.
  • A tool gets a page of its own only when its documentation says enough to quote; the others are rows here with the official page we read, or a note that none could be read.
  • Google Veo is covered on the Gemini, Imagen and Veo page.

Sources

Each entry links to its own sources, with the date each was last checked.

    Page reviewed . Proofcan is a technical tool, not legal advice, and does not guarantee compliance.