AI People Wiki

Builders & open source

Sven Gowal

Co-led Google DeepMind's SynthID project; co-author of the SynthID Text and SynthID Bio papers

Sven Gowal co-led Google DeepMind's SynthID project for watermarking and identifying AI-generated content with Pushmeet Kohli, according to the 2023 SynthID launch post. Gowal is a co-author of the Nature papers behind SynthID Text, Google DeepMind's open-source text watermarking library, and SynthID Bio, its watermarking tool for AI-generated proteins, and is first author of the SynthID-Image paper. Gowal's earlier research covers adversarial robustness and verification of neural networks.12345

At a glance

Current role
None confirmed in the sources
Work listed
3
Sources cited
8
Profile reviewed

SynthID

Google DeepMind's August 2023 post announcing a beta of SynthID, which embeds an imperceptible watermark in image pixels, was written by Gowal and Pushmeet Kohli and states that the two led the project. Gowal is first author of the 2025 SynthID-Image paper on image watermarking at internet scale.14

Gowal is also a co-author of the Nature papers for SynthID Text (2024), on watermarking the outputs of large language models, and SynthID Bio (2026), on function-preserving watermarking of AI-generated proteins; both are released as Apache-2.0 repositories under Google DeepMind's GitHub organization.23

Research

Gowal's earlier papers address robustness and verification of neural networks, including interval bound propagation for training verifiably robust models (2018), the limits of adversarial training against norm-bounded examples (2020) and using generated data to improve robustness (2021).567

Work and contributions

  • SynthID Text Open-source code

    Co-author of "Scalable watermarking for identifying large language model outputs" (Nature, 2024), the paper the SynthID Text repository asks users to cite.2

  • SynthID Bio Open-source code

    Co-author of "Function-preserving watermarking of AI-generated proteins" (Nature, 2026), the paper cited by the SynthID Bio repository.3

  • SynthID-Image Published research

    First author of "SynthID-Image: Image watermarking at internet scale" (2025).4

Timeline

  1. Posted "On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models" as first author.5
  2. Posted "Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examples" as first author.6
  3. Posted "Improving Robustness using Generated Data" as first author.7
  4. Co-authored Google DeepMind's announcement of the SynthID beta, a project Gowal co-led.1
  5. Co-authored the Nature paper on scalable watermarking of LLM outputs behind SynthID Text.2
  6. Posted "SynthID-Image, Image watermarking at internet scale" as first author.4
  7. Co-authored the Nature paper on function-preserving watermarking of AI-generated proteins behind SynthID Bio.3

Documented roles

The sources do not confirm a current role; these are the roles they document.

  • Co-led the SynthID project (2023 launch post); listed with a Google DeepMind affiliation on the SynthID Bio paper (Nature, September 2026), Google DeepMind18

Companies in their history

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

8 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026

  1. Identifying AI-generated images with SynthIDGoogle DeepMind · Announcement · published Aug 29, 2023 · accessed Oct 22, 2026
  2. google-deepmind/synthid-text README (citation)Google DeepMind (GitHub repository) · Repository · accessed Oct 22, 2026
  3. google-deepmind/synthidbio README (citation)Google DeepMind (GitHub repository) · Repository · accessed Oct 22, 2026
  4. SynthID-Image: Image watermarking at internet scale (arXiv 2510.09263)arXiv · Paper · published Oct 10, 2025 · accessed Oct 22, 2026
  5. On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models (arXiv 1810.12715)arXiv · Paper · published Oct 30, 2018 · accessed Oct 22, 2026
  6. Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examples (arXiv 2010.03593)arXiv · Paper · published Oct 7, 2020 · accessed Oct 22, 2026
  7. Improving Robustness using Generated Data (arXiv 2110.09468)arXiv · Paper · published Oct 18, 2021 · accessed Oct 22, 2026
  8. Function-preserving watermarking of AI-generated proteins (Nature)Nature · Paper · published Sep 30, 2026 · accessed Oct 22, 2026