Veo (Google) and GDPR / AI Act: SynthID, deepfakes and image rights
In brief
- Veo (Google) produces realistic videos from text, with SynthID watermarking (invisible watermark). Generative media issues: do not represent real people without consent, disclose deepfakes (Article 50), frame transfers outside the EU (CLOUD Act). SynthID facilitates technical compliance but does not exempt from visible disclosure on the deployer side. Check the offering and data residency.
Overview and data processed
Veo is Google's video generation model (via its offerings and Gemini): it produces realistic videos from text descriptions, with integrated SynthID watermarking.
Veo processes prompts and any images provided, and can generate identifiable people. US publisher: exposure to the CLOUD Act; check the offering and data residency.
Risks for your organisation
The risks: representation of real people without consent, misleading deepfakes, transfers outside the EU and copyright.
Main GDPR risks:
- Representation of real people without consent (image rights, voice as personal data).
- Transfer outside the EU (CLOUD Act); prompts containing personal data.
- Origin of training data (scraping prohibited); copyright.
- Misleading content (disinformation).
GDPR obligations
- Prohibit the representation of real people without documented consent.
- Check the DPA and data residency; frame transfers (SCCs).
- Do not enter unnecessary personal data; record in the record of processing activities; DPIA if high-risk use.
- Inform the people represented.
- Involve the DPO; train teams. No DPO? See our outsourced DPO.
AI Act obligations
Article 50 (applicable 2 August 2026, technical deadline 2 December 2026): the provider marks technically; the deployer who produces a deepfake must disclose clearly. Lighter regime for artistic/satirical content. GPAI; AI literacy. Prohibition (Article 5): untargeted facial scraping prohibited. See the AI Act guide.
AI Act risks: undisclosed deepfakes (Article 50); SynthID watermark retained but visible disclosure omitted; misleading content on matters of public interest.
Cybersecurity and best practices
Cyber risks: impersonation via generated video, disinformation, transfer outside the EU; NIS2 depending on size.
Best practices: prohibit the representation of real people without consent; retain SynthID watermark and add visible disclosure; editorial control; frame transfers; AI charter (prohibited uses); record of processing activities, training.
Checklist:
- No representation of real people without consent, SynthID watermark retained + visible disclosure
- DPA + data residency verified, transfers framed (SCCs), no unnecessary personal data in prompts
- DPIA if high-risk use, AI charter (prohibited uses), editorial control, teams trained
Sanctions, conclusion and recommendations
Sanctions: GDPR (up to €20m or 4%); AI Act (€15m or 3% for Article 50, €35m or 7% for prohibited practices); civil and criminal — image rights, articles 226-8 and 226-8-1 of the Criminal Code.
Recommendation: Veo benefits from native watermarking (SynthID) facilitating technical compliance, but the deployer remains responsible for visible disclosure of deepfakes and respect for image rights. Prohibit the representation of real people without consent, frame transfers and control before distribution. Compare with Sora and Runway. See the AI tool comparison.
Take action
Assess the level of GDPR and AI Act risk of your use of Veo with the Data Comply One self-assessment.
Discover the AI Act software →Test my compliance for free