Prompt & Done · CREDENTIALS pack

Does this image carry Content Credentials? The checker from the reel.

Content Credentials (C2PA) are signed metadata inside an image file: who issued it, which tool made it, whether it is AI-generated. They are optional, and they are easy to lose.

1 The checker

python3.11 -m venv v && v/bin/pip install c2pa-python
v/bin/python check.py your-image.jpg
import sys, json, c2pa

for path in sys.argv[1:]:
    try:
        data = json.loads(c2pa.Reader(path).json())
        m = data["manifests"][data["active_manifest"]]
        acts = [a for a in m["assertions"] if a["label"].startswith("c2pa.actions")]
        act = acts[0]["data"]["actions"][0] if acts else {}
        sig = m["signature_info"]
        print(path, "CREDENTIALS FOUND", sig.get("issuer"), act.get("softwareAgent"), act.get("action"), str(act.get("digitalSourceType", "")).split("/")[-1], str(sig.get("time"))[:10], data.get("validation_state"))
    except Exception as e:
        print(path, "NO CREDENTIALS FOUND (that proves nothing about who made it)")

Tested 26 Oct 2026 with c2pa-python 0.38.1 on macOS. Reads a file and prints what is signed inside it; it does not detect AI by itself.

2 What we ran in the reel (26 Oct 2026)

One example, one tool. Not every export or screenshot method behaves the same, and some platforms keep or re-attach credentials.

3 What OpenAI says

OpenAI's developer docs say its images can carry C2PA Content Credentials and a SynthID watermark, and that you can check a file at openai.com/verify. They also say: "Compression, cropping, screenshots, metadata removal, and format conversions can erase or weaken a signal." And on a "not detected" result: "Treat not_detected as an absence of detected evidence." It "doesn't rule out AI-generated or AI-modified content."

OpenAI developer docs: Content provenance

4 Links