Secret Watermark Labels AI-Designed Proteins

Researchers have developed a novel method to embed a "secret watermark" within proteins designed by artificial intelligence (AI) tools, according to a study published online in Nature on September 30, 2026. This digital marker aims to identify proteins that have been generated by AI, such as those produced by protein-design software like AlphaFold, a widely used AI system developed by DeepMind for predicting protein structures. The watermark is designed to be imperceptible to biological systems but detectable through specific analytical techniques.

The technique involves subtly altering the amino acid sequence of the protein in a way that does not affect its intended function or structure. These alterations are strategically placed and follow a specific pattern that acts as a unique identifier. When a protein is suspected of being AI-designed, scientists can use a corresponding decoding algorithm to scan for this pattern. If the pattern is detected, it confirms that the protein was likely generated or significantly modified by an AI model. This development is particularly relevant in the rapidly advancing field of protein engineering, where AI is increasingly used to design novel proteins with specific therapeutic or industrial applications.

However, the researchers also noted a significant limitation: the watermark can be erased. If a protein is intentionally modified or if the AI design process is not carefully controlled, the watermark might be inadvertently removed or altered, making it undetectable. This presents a challenge for the long-term reliability of the watermark as a definitive identifier. The study highlights the ongoing need for robust methods to distinguish between naturally occurring biological molecules and those created or manipulated by AI, especially as AI's capabilities in biological design continue to expand.

The ability to watermark AI-designed proteins could have several implications. In scientific research, it could help researchers track the origin of novel protein designs and ensure proper attribution. In the pharmaceutical industry, where AI is being used to discover and design new drugs, it could provide a layer of verification. The development also touches upon broader discussions about the ethical implications of AI in science and the need for transparency in AI-driven discoveries. The researchers are continuing to explore ways to make the watermark more resilient to tampering and to develop complementary methods for identifying AI-generated biological entities.