Artificial Intelligence
Google DeepMind unveils SynthID Bio
Google DeepMind published research on Wednesday describing SynthID Bio, a watermarking approach for AI-designed proteins intended to identify trusted sequences and subject others to closer scrutiny.
- Google DeepMind published a research paper on Wednesday about SynthID Bio.
- The paper said the system could watermark protein sequences without affecting protein function.
- The article said software used to detect DNA sequences linked to threatening proteins did not identify AI-designed proteins.
- The work was based on Google's SynthID technology, which adds subtle watermarks to AI-generated digital material.
The article said AI tools were contributing to security threats and that biosecurity was one area where people appeared to be ahead of the problem. It also said protein-design tools had produced enzymes that could break down plastics or block venom proteins.
It said the same tools could also be used to make toxins or change the behaviour of viral proteins. Existing software for spotting DNA sequences associated with dangerous proteins did not recognise AI-designed proteins, because they had not been characterised well enough.
On Wednesday, the DeepMind team at Google published a paper proposing protein watermarking. The system marked protein sequences without changing their function, so proteins made by trusted researchers could be identified and other sequences examined more closely.
The paper said the method drew on Google's SynthID, which places subtle watermarks in AI-generated digital material by biasing generation choices. It said the watermark could not be detected without the encoding and was designed to survive basic exporting and resizing.
Named in this story
Companies
- was the company behind the SynthID technology used as the basis for the work
Organisations
- Google DeepMind
- published a research paper on protein watermarking
Products and systems
- SynthID Bio
- was the proposed watermarking method for AI-designed proteins
- SynthID
- was the existing watermarking system that the protein work drew on
How the source tells it
The register starts with modest alarm about AI biosecurity and shifts into cautious optimism about a proposed watermarking fix.
- alarm an opening generalisation about AI tools causing near-daily security threats
- cautious optimism a contrastive turn to a potential solution that frames biosecurity as ahead of the problem