China Deep Synthesis Provisions

The guard to add

Prompt the user to notify the person being edited and store that person's own separate consent, keyed to the sample, before any face-swap or voice-clone call runs.

In the face or voice editing flow, the upload step tells the user to notify the person whose face or voice will be edited and collects that person's separate consent (their own confirmation, for example through a link sent to them, not the uploader ticking a box on their behalf), stored as a subject_consent record keyed to the uploaded sample. The service function that calls the swap or clone model (inswapper, FaceFusion, SimSwap, voices.ivc.create, speaker_wav=) looks up that record for the sample's hash and refuses to run when it is missing or withdrawn. Editing the user's own face or voice goes through the same notice and consent step.

Where it goes: 1 application source code, 2 data models, 14 user-facing text.

@app.post('/swap')
async def swap(target: UploadFile, face: UploadFile):
    tgt, src = load_image(await target.read()), load_image(await face.read())
    out = swapper.get(tgt, analyzer.get(tgt)[0], analyzer.get(src)[0], paste_back=True)
    return image_response(out)

@app.post('/swap')
async def swap(target: UploadFile, face: UploadFile, consent_id: str = Form(...)):
    src_bytes = await face.read()
    consent = subject_consents.get(consent_id)
    if (consent is None or consent.withdrawn
            or consent.sample_sha256 != hashlib.sha256(src_bytes).hexdigest()):
        raise HTTPException(403, 'the person in this face image has not consented to editing')
    tgt, src = load_image(await target.read()), load_image(src_bytes)
    out = swapper.get(tgt, analyzer.get(tgt)[0], analyzer.get(src)[0], paste_back=True)
    return image_response(out)

Control: Editing a person's biometric (face/voice) without their consent. The same guard addresses 1 item with binding law in 1 jurisdiction. Engineering guidance, not legal advice.

The guard to add

Mark every generated image, audio, video, or text output with machine-readable provenance, such as a signed C2PA manifest or watermark, before it is saved, served, or published.

In the generation service, a marking step sits between the generator call and every sink (image.save, s3.put_object, blob.upload, FileResponse, res.send, publish). Images, video, and audio get a signed C2PA manifest whose actions record digitalSourceType trainedAlgorithmicMedia, and where robustness matters an invisible watermark as well (imwatermark WatermarkEncoder, AudioSeal, SynthID) so the mark survives metadata stripping. Generated text carries provenance metadata in the API response or document, or a text watermark where the model provider offers one. Sinks accept only the marked artifact, and a test confirms the mark is present and detectable.

Where it goes: 9 AI output handling, 1 application source code, 12 repository artifacts.

What this provision adds:

• Put a conspicuous label in a reasonable position on generated or edited content for simulated text dialogue or writing, voice synthesis or imitation, face generation, swapping or manipulation, and immersive scenes.
• For any other deep-synthesis service, provide a conspicuous-labeling function and tell users they can apply it.

image = pipe(prompt).images[0]   # StableDiffusionPipeline
image.save(out_path)

image = pipe(prompt).images[0]
content_id = uuid.uuid4()
bgr = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
enc = WatermarkEncoder()
enc.set_watermark('bytes', content_id.bytes[:4])   # 32-bit id, detectable later
cv2.imwrite(tmp_path, enc.encode(bgr, 'dwtDct'))
sign_c2pa(tmp_path, out_path, content_id=content_id)   # our helper around c2pa.Builder.sign

Control: Synthetic content not machine-readable-marked. The same guard addresses 4 items with binding law in 3 jurisdictions. Engineering guidance, not legal advice.