{
  "id": 11217657,
  "title": "Google DeepMind Adds Watermarks to AI-Designed Proteins Without Breaking Them",
  "url": "https://urgent.news/2026/10/01/google-deepmind-adds-watermarks-to-ai-designed-proteins-without",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-01T14:25:31.000Z",
  "source": {
    "name": "TechRepublic",
    "slug": "techrepublic",
    "url": "https://www.techrepublic.com/article/news-google-synthid-bio-ai-protein-watermark/"
  },
  "original_language": "en",
  "account": "Google DeepMind has introduced SynthID Bio, a technology that embeds invisible watermarks into AI-designed proteins while maintaining their functionality. This development addresses the challenge of identifying the origins of AI-generated biological sequences, which can be difficult to trace using traditional screening methods. The watermarking process subtly alters the selection of amino acids in proteins, allowing a specialized verification system to detect the presence of a SynthID Bio-enabled design. In laboratory tests, proteins watermarked through SynthID Bio retained their function and predicted 3D structures remained largely accurate, preserving the effectiveness of the tagged sequences. DeepMind has integrated the technology into various platforms, including AlphaProteo and ProteinMPNN, to design protein binders, which are molecules engineered to attach to specific targets. Early tests with VEGF-A, the SARS-CoV-2 spike protein's receptor-binding domain, and PD-L1 showed that watermarked proteins performed comparably to unmarked ones in terms of hit rate, binding affinity, and natural sequence diversity. The potential applications of SynthID Bio include helping DNA synthesis companies differentiate AI-generated sequences from unfamiliar natural proteins, enhancing the security of DNA synthesis providers by focusing scrutiny on sequences without recognizable provenance, and safeguarding scientific databases such as the Protein Data Bank, UniProt, and GenBank from mislabeled synthetic entries. Despite its promising results, SynthID Bio is not a complete solution to biological security risks, as making watermarks resistant to intentional removal remains a significant challenge. Further research and standardization efforts are required for wider adoption by protein designers, synthesis providers, and database managers.",
  "summary": "Google DeepMind’s SynthID Bio embeds detectable watermarks in AI-designed proteins while preserving their function in early laboratory tests. The post Google DeepMind Adds Watermarks to AI-Designed Proteins Without Breaking Them appeared first on TechRepublic .",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Times of India",
        "title": "Google DeepMind watermarks AI proteins as Demis Hassabis warns of biosecurity risks",
        "url": "https://urgent.news/2026/10/01/google-deepmind-watermarks-ai-proteins-as-demis-hassabis-warns-of",
        "published": "2026-10-01T12:20:56.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}