
Model Output Governance: Watermarking and Copyright
Dealing with the legal grey areas. Understanding SynthID (Watermarking) and managing copyright risks in generated content.
Model Output Governance
When Gemini writes a poem, who owns it?
Copyright
Current US law states AI Generated content is not copyrightable. You cannot file a DMCA takedown if someone steals your AI-generated blog post. (This is legal advice lite; consult a lawyer).
- Risk: If Gemini inadvertently reproduces a copyrighted logic or character (e.g., Mickey Mouse), you could be liable for infringement. (Google provides "Indemnification" for paid users on Vertex AI).
Watermarking (SynthID)
Google embeds an invisible digital watermark (SynthID) into the pixels of generated images and the tokens of generated text.
- Purpose: Allows detectors to prove "This came from Google AI."
- Action: You don't need to do anything; it's automatic. But be aware it exists transparency.
Usage Policies
Google prohibits using Gemini for:
- Generating malware.
- Generating Non-Consensual Sexual Imagery (CSAM).
- High-risk medical advice without disclaimers.
Summary
Read the AUP (Acceptable Use Policy). If you violate it, your API access will be revoked instantly.
In the next lesson, we discuss Safe Sharing.