The United States and China rarely appear to have much common ground on artificial intelligence. Yet one proposal emerging from the growing debate over AI-enabled biological risks offers something unusual: a safeguard that both countries could pursue without first resolving their deeper strategic disagreements.
As China Talk argues in the analysis on which this article is based, the screening of synthetic DNA and RNA orders could provide a practical starting point for US-China cooperation on biological risk.
The proposal has attracted support from an unusually broad group of figures in artificial intelligence and biosecurity. In a June open letter, Sam Altman, Dario Amodei, Demis Hassabis and a long list of AI and biosecurity figures from both the right and left argued that the United States should require nucleic acid synthesis companies to screen DNA and RNA orders before fulfilling them.
The underlying concern is straightforward. Artificial intelligence may make it easier to design dangerous pathogens, but digital designs ultimately have to enter the physical world. Nucleic acid synthesis providers represent one of the few points at which biological information can become physical biological material.
Screening those orders therefore creates a potential security chokepoint.
Nucleic acid synthesis screening is essentially a security check for custom DNA and RNA orders. When an order is placed, a provider can compare the sequence against databases of dangerous pathogens and toxins, assess whether the customer appears suspicious, and decide whether the order should be fulfilled, delayed or rejected.
The purpose is not to scrutinise every biological experiment. It is to prevent dangerous genetic material from being supplied without appropriate scrutiny.
The importance of that distinction is growing as AI-enabled biological design advances. A researcher could already seek to synthesise genetic material associated with dangerous pathogens whose sequences have been published. The concern is that increasingly capable AI systems could make it easier not merely to reproduce known pathogens but to design new biological sequences with dangerous characteristics.
Many major synthesis companies already screen orders voluntarily. But participation is not universal, and enforcement remains difficult. A single poorly regulated supplier can create a loophole through which an otherwise extensive screening system can be bypassed.
That asymmetry is central to the biosecurity problem. A system may perform successfully almost every time and still fail catastrophically if one capable bad actor discovers a gap or if one legitimate laboratory makes a serious mistake. The large number of harmless orders does not compensate for a single dangerous one.
AI model safeguards have a role, but they are not sufficient on their own. Model-level protections can sometimes be circumvented, while efforts to reduce biological risks at the model level can result in blunt measures that restrict legitimate research. Screening the physical synthesis supply chain approaches the problem at a different point: before a dangerous sequence becomes a shipped product.
The United States has already moved towards mandatory screening, although the process has been unsettled. Last year, the Trump administration cancelled the Biden-era nucleic-acid synthesis screening framework and promised to replace it with something broader and more enforceable. According to the China Talk analysis, biosecurity subsequently received less attention, but renewed concern about frontier AI risks has increased recognition that biological risks associated with advanced AI may not be purely theoretical.
There are also legislative efforts under way. Cotton and Klobuchar have a Senate bill that would require Commerce and OSTP to act on synthesis screening, while Pfluger and Houlahan have introduced a narrower House bill.
But an American system alone cannot secure a global biological supply chain.
China accounts for approximately 34% of the world’s DNA synthesis providers, according to the material cited by China Talk. Its AI ecosystem is also more open-source, creating additional difficulties for model-level safeguards. The analysis argues that Chinese AI models have shown weaker safeguards against biological misuse, while China’s simultaneous push towards open-source AI and biotechnology increases the importance of downstream controls.
China, however, is not starting from nothing. It does not yet have a mandatory national order-screening regime, but some major China-linked synthesis providers participate in voluntary screening systems. BGI Bio-Solutions joined the International Gene Synthesis Consortium in 2017, while GenScript is also a member. Chinese companies and experts have additionally worked with the International Biosecurity and Biosafety Initiative for Science, or IBBIS, which seeks to make synthesis screening easier and more standardised across countries.
Chinese researchers have also increasingly raised synthesis screening as an issue.
That creates a potentially unusual basis for cooperation. Both countries have reasons to reduce the possibility of a pandemic emerging within their borders, while neither necessarily has to surrender a core strategic interest to achieve that objective.
This is where the distinction between contingent and demonstrated cooperation becomes important. Contingent cooperation depends on both sides acting simultaneously and honouring reciprocal commitments. Demonstrated cooperation instead involves each country independently taking actions that advance an objective it already has reason to pursue, while providing evidence that meaningful action has occurred.
Synthesis screening fits the second model.
Washington could require American providers to screen orders, while Beijing could require Chinese providers to do the same. Each government could demonstrate domestically that it was addressing AI-enabled biological risks without making its actions dependent upon concessions from the other.
For upcoming US-China AI talks, that makes synthesis screening unusually practical. It does not require either side to make AI systems weaker, abandon open-source development or accept intrusive verification. Nor does it necessarily impose significant costs on biotechnology. Ordering DNA takes days, while developing a drug can take years, with failed constructs, cell work, animal studies, clinical trials, manufacturing and regulatory review representing much larger bottlenecks.
Screening technology itself is also becoming cheaper. IBBIS’s Common Mechanism provides a baseline tool for sequence screening, while advances in AI-assisted customer verification could reduce the cost of checking customers.
Companies representing roughly 80% of global synthesis capacity already screen voluntarily. Mandatory rules would therefore primarily raise standards among the remaining providers rather than impose an entirely new system on the industry.
The economic argument is reinforced by the potential consequences of another pandemic. The material cited by China Talk notes that, at its peak, COVID was costing the US economy roughly $26 billion a day. A UK analysis cited in the source found that every £1 spent on synthetic nucleic-acid screening could return £3.50 in averted losses.
Yet screening is not a complete solution. Benchtop synthesizers, fragmented orders and overseas suppliers can weaken the system. Split ordering, in which different fragments are obtained from different providers or countries, could eventually require more direct information-sharing. Benchtop synthesis presents another challenge.
That is precisely why the proposal is framed as a beginning rather than an endpoint.
Rather than immediately placing the issue inside a grand US-China biosecurity treaty, the two countries could establish a shared baseline covering which DNA and RNA orders should be screened, what sequence lengths should qualify, which pathogen or toxin categories should trigger review, and what customer information providers should verify.
They could then begin exchanging aggregate information about screening coverage and progress. More sensitive information about flagged customers could initially remain confidential, potentially with trusted organisations such as SecureDNA, IBBIS, IGSC, NTI and SAIF helping develop standards and technical dialogue.
China Talk argues that routing the entire effort through institutions such as the Biological Weapons Convention or the World Health Organisation could make cooperation more politically difficult by introducing disputes over verification, sovereignty and accusations of non-compliance.
The more modest approach is to begin bilaterally, with each side demonstrating that it can act.
For Washington and Beijing, the significance of such an arrangement would extend beyond DNA screening. The immediate objective would be to close one vulnerable connection between digital biological knowledge and physical biological capability. The longer-term test would be whether two strategic competitors can establish trust through practical, independently demonstrated action.
In a field where the consequences of failure can be vastly greater than the cost of prevention, that may be one of the few areas of US-China AI cooperation where competition and common interest can coexist.
