Google figures out how to watermark AI-designed proteins

Google has developed a watermarking system for AI-designed proteins that embeds identifying marks into the protein sequences without affecting their function. The watermark works by influencing which amino acids the AI protein design tool selects at each step, distributing the signal throughout the entire protein so it cannot be removed. This matters for biosecurity because DNA synthesizers currently cannot distinguish between AI-designed proteins created by trusted researchers and those created by bad actors, making it difficult to screen orders for potentially dangerous proteins like toxins or altered viral proteins. The watermarking system allows DNA synthesizers to verify that unknown proteins come from trusted sources, helping them focus their security screening efforts on designs that lack these trusted watermarks.