The federal framework intended to prevent the manufacture of dangerous genetic sequences on demand has not been revised or replaced more than a year after an executive order directed agencies to do so. That gap is drawing renewed attention now that generative AI has produced complete, functional viral genomes for the first time.
A Congressional Research Service assessment updated in July found that, as of that month, it could not determine whether the 2024 Framework for Nucleic Acid Synthesis Screening had been revised or replaced, despite a 90-day deadline set by Executive Order 14292 in May 2025. The federal screening framework page maintained by HHS still carries a notice saying it will be updated once a revised framework is available. The screening requirement established by the 2024 framework applies as a condition of federal research funding rather than as a universal legal mandate.
The practical meaning is narrow but real. The main technical chokepoint between a digital sequence and a physical organism is the company that synthesizes DNA to order. Screening at that step is voluntary for providers outside the federal funding requirement, and the policy meant to strengthen it is in limbo.
The Result That Reopened the Question
Researchers at Stanford University and the Arc Institute reported in Science that genome language models generated roughly 700,000 candidate bacteriophage genomes using the phage ΦX174 as a design template. The team selected 302 for synthesis, successfully built 285, and 16 became working viruses that infected and killed E. coli. Some replicated faster than the natural reference phage, and a cocktail of the designed phages rapidly overcame resistance in three ΦX174-resistant bacterial strains that the natural phage alone could not.
MedicalDaily reported on that study and the biosecurity warnings that accompanied it when it was published. The researchers took deliberate precautions. Sequences from viruses capable of infecting complex organisms were excluded from the models' training data, and the work used a phage that cannot infect people. The Arc Institute has publicly detailed its design approach and says the work followed biosafety protocols that exceeded standard requirements.
The genome they built is also small. ΦX174 runs 5,386 nucleotides and encodes 11 genes, and the institute chose it partly because it sits near the upper limit of what current DNA synthesis costs make practical. The SARS-CoV-2 genome is more than five times longer, and complex pathogen traits such as transmissibility and immune escape are not reliably predictable with current models. Nobody has demonstrated an AI system designing a human pathogen, and that distinction should not be blurred.
The hit rate is also worth stating plainly. Of the 285 genomes successfully built, 16 produced viable phages, a success rate of under 6 percent. Nine of the 16 matched the sequences the models produced exactly, while seven picked up additional mutations after being introduced into bacteria. This is a demonstration of feasibility at genome scale, not a reliable manufacturing process.
The Screening Layer Policymakers Keep Returning To
Sequence screening asks providers of synthesized DNA and manufacturers of benchtop synthesis machines to check orders against sequences of concern and verify that the customer is legitimate before shipping. Synthesis companies formed the International Gene Synthesis Consortium in 2009 to do this voluntarily.
In June, a coalition of life sciences researchers, national security experts, AI executives, and leaders of DNA synthesis companies published an open letter urging Congress to make screening of synthetic nucleic acid orders and of the equipment used to produce them mandatory. The letter also called for providers to record orders and sequence data so that any threat evading initial screening could be traced back to its source, including cases where individual sequences would not raise concern in isolation. Notably, the responsibilities it proposes fall on synthesis companies rather than on AI model developers.
A Senate bill introduced this year, the Biosecurity Modernization and Innovation Act of 2026, would require the Secretary of Commerce to establish regulations for nucleic acid synthesis security, including a maintained list of sequences of concern and screening protocols that covered providers would have to implement. It has not become law, and several earlier gene synthesis bills did not advance.
A parallel gap exists in what gets screened at all. The 2024 framework defined sequences of concern at 200 or more nucleotides, with a scheduled reduction of that threshold, meaning shorter oligonucleotides have often been screened lightly or not at all, even though standard laboratory techniques can assemble them into longer sequences. Benchtop synthesis machines that let a laboratory produce DNA without placing an outside order sit outside the order-screening model entirely, which is why several proposals would push screening software into the devices themselves.
There is a technical wrinkle that matters here. Homology-based screening works by matching an order against known dangerous sequences. Generative design can produce sequences with low similarity to anything in existing databases, as demonstrated by the Stanford team. A screening method built on resemblance to known threats is a weaker filter against sequences that do not resemble anything on file.
The Household Stake Runs Through Treatment Access
This is a policy and research story, not a public health alert. No one is being exposed to anything. The phages involved infect bacteria, not people, and no illness or release is associated with this work.
The household relevance runs in the other direction, toward treatment. Antibiotic-resistant infections cause more than 2.8 million illnesses in the United States each year, and phage therapy is one of the few alternative approaches with a real clinical track record. It is also almost entirely inaccessible. As MedicalDaily has reported, no phage product has been approved for human use in the United States, meaning it is not covered by insurance and is available only on a case-by-case basis through FDA expanded access, arranged by an infectious disease physician.
An AI system that can design phages faster does not shorten that pathway. Regulatory approval, manufacturing standards, and clinical trials sit between a laboratory result and a treatment anyone can receive. Patients or families facing a resistant infection should discuss phage therapy with an infectious disease specialist rather than pursuing products marketed directly to consumers, which have not been evaluated for safety or effectiveness.
Several things remain unresolved. Federal agencies have not published a replacement screening framework, the Senate bill has not advanced, and it is not clear how the eventual framework will treat sequences that no existing database would flag. MedicalDaily will report when a replacement framework is issued.
Key Questions Answered
What is the new development here? A Congressional Research Service assessment updated in July found the 2024 federal framework for nucleic acid synthesis screening had not been revised or replaced, despite a 90-day deadline set in an executive order signed in May 2025.
Why does DNA synthesis screening matter? It is the main checkpoint between a digital genetic sequence and a physical organism. Providers check orders against sequences of concern and verify customers before shipping.
Did AI create something dangerous? No. The designed viruses infect E. coli bacteria and cannot infect people. Sequences from viruses that infect complex organisms were excluded from the models' training data.
Could the same approach design a human pathogen? Not demonstrated. The phage genome is 5,386 nucleotides, less than a fifth the length of the SARS-CoV-2 genome, and traits like transmissibility are not reliably predictable with current models.
Why is screening harder now? Existing methods match orders against known dangerous sequences. Generatively designed genomes can show low similarity to anything in existing databases.
Is phage therapy available to patients? Not as an approved treatment. No phage product is FDA-approved for human use in the United States, so it is not covered by insurance and is accessed on a case-by-case basis through expanded access.
What should a patient with a resistant infection do? Talk with an infectious disease specialist about whether expanded access is an option. Phage products sold directly to consumers are not evaluated for safety or effectiveness.
