Synthesia and GDPR / AI Act: AI avatars, image rights and labelling

Synthesia and GDPR / AI Act: AI avatars, image rights and labelling

In brief

  • Synthesia (British publisher) generates videos with AI avatars. Two issues: the right to image and voice ('custom' avatars created from real people = written, specific, revocable consent) and marking/disclosure of generated content (Article 50, applicable 2 August 2026). The United Kingdom benefits from an EU adequacy decision. Verify the origin of training data (untargeted facial scraping prohibited).

Overview and data processed

Synthesia is a British AI video generation platform: from text, it produces an animated video featuring an avatar with a synthetic voice. It offers catalogue avatars and personalised avatars (cloning of an employee).

Synthesia processes the text provided and, for personalised avatars, the image and voice of the cloned person (potentially biometric data). The videos may represent identifiable individuals. British publisher: transfers covered by UK adequacy; check hosting and processors.

The risks for your business

The main risk is the right to image and voice: creating an avatar from an employee without adequate consent, or making them deliver binding statements without a framework. There is also the obligation of transparency on generated videos.

Main GDPR risks:

  • Reproducing the face or voice of a real person requires their consent (right to image, voice as personal data).
  • Custom avatar: written, specific and revocable consent; purposes and duration.
  • Transfers (UK adequacy) and processors to verify.
  • Origin of training data (facial scraping prohibited).

GDPR obligations

  1. Obtain written and specific consent (image + voice) for any personalised avatar.
  2. Frame uses and revocation; inform the individuals represented.
  3. Verify the DPA, hosting and processors.
  4. Record in the record of processing activities; DPIA if large-scale biometric processing.
  5. Involve the DPO; train communication/training teams. No DPO? See our outsourced DPO.

AI Act obligations

Article 50 (applicable 2 August 2026, technical deadline 2 December 2026): the provider marks content technically (watermark, C2PA/IPTC metadata) and the deployer producing a deepfake must clearly disclose that it has been generated/manipulated by AI — a mention buried in the T&Cs is insufficient. Lighter regime (minimal disclosure) for manifestly artistic/satirical works. GPAI models; AI literacy (Article 4). Prohibition (Article 5): untargeted facial scraping is prohibited — verify the origin of training data. See the AI Act guide.

AI Act risks: generated videos not marked/not disclosed (Article 50); deepfakes of real people without disclosure or consent; data from facial scraping (prohibited).

Cybersecurity and best practices

Cyber risks: impersonation via avatars (CEO fraud in 'video'), distribution of compromising videos, access to personalised avatars to protect; NIS 2 depending on size.

Best practices: written consent for each personalised avatar (register); systematic marking and disclosure; exploit native marking and verify the origin of data; AI charter, training, anti-impersonation procedure (double validation).

Checklist:

  • Written consent (image + voice) for personalised avatars, uses and revocation framed, individuals informed
  • Marking/disclosure of videos (Article 50), DPA + hosting/processors verified, origin of data verified
  • DPIA if large-scale biometric processing, record of processing activities up to date, teams trained

Sanctions, conclusion and recommendations

Sanctions: GDPR (up to €20m or 4% of global turnover); AI Act (up to €15m or 3% for a breach of Article 50 transparency, €35m or 7% for prohibited practices); civil and criminal — infringement of the right to image, Article 226-8 of the Criminal Code (non-consented editing), Article 226-8-1 (sexual deepfakes).

Recommendation: Synthesia is a powerful tool, but avatars touch on the right to image and voice: require written and specific consent, mark and disclose generated videos, verify the origin of data and put in place an anti-impersonation procedure. Compare with HeyGen and video generators Sora, Veo. See the AI tools comparison.

Take action

Assess the GDPR and AI Act risk level of your use of Synthesia with the Data Comply One self-assessment.

Discover the AI Act software →Test my compliance for free
Compare this tool to the 70 AIs analysed — hosting, training, DPA and verdict.View the full comparison →

Frequently asked questions

Is Synthesia GDPR compliant?

It can be, provided you obtain consent for personalised avatars (image and voice), frame uses, verify hosting/processors and mark/disclose generated videos.

Is consent required to create an avatar of an employee?

Yes: reproducing the image and voice requires written, specific and revocable consent (right to image, Article 9 of the Civil Code, and voice as personal data).

Must I disclose that my video is AI-generated?

Yes: Article 50 of the AI Act requires marking and, for a deepfake, disclosing that the content has been generated/manipulated by AI. A mention in the T&Cs is not sufficient.

Is the United Kingdom an adequate country?

Yes: it benefits from an EU adequacy decision; however, verify actual hosting and processors.