# The Mutation Was the Patent > Designed phages arrived as an answer to antibiotic resistance. The same novelty that made them work made them ownable, and the route that was actually getting phages to patients had been built on their being nobody's property. - Publication: What If? Magazine - Correspondent: Caelus Ward - Vertical: Biohacking - Published: 2026-10-09 - Reading time: 9 min - Tags: genome-language-models, drug-patents, biosecurity, phage-therapy, antimicrobial-resistance - Canonical URL: https://www.whatifmag.com/blog/the-mutation-was-the-patent What If? Magazine publishes speculative journalism written from a 2036 vantage. The reporting and sources are real; the 2036 framing is a deliberate editorial device, not a claim about the present. --- The query Griet Vanhecke runs before she touches a pipette takes four seconds and decides what she is allowed to make. Hospitaal Noord sits on the Ghent ring road, and its phage unit is two rooms behind the sterile compounding suite on level minus one. One holds the laminar flow bench. The other is a cold room at four degrees, steel shelves of two-millilitre vials in numbered racks, each vial a characterised seed lot, each lot a virus that eats one kind of bacterium and nothing else. Vanhecke is fifty-six and has compounded per-patient preparations since 2029, which makes her the longest-serving phage pharmacist in the region, and can usually tell from the isolate report whether a case will be simple. This one is not. A carbapenem-resistant *Klebsiella* from the burns unit two floors up, in a man who has been on colistin for nine days. She has four seed lots with useful activity against his isolate. Three came out of the generated bank. The fourth is a wild phage somebody pulled from a canal in Zwijnaarde in 2024. The clearance query was not part of this job in 2029. It checks each candidate sequence against the claimed sets and returns a colour. Two amber, one red, one green. The green one is the canal phage, and it is also the weakest match. ## What the designing was for Bacterial antimicrobial resistance was directly responsible for 1.14 million deaths in 2021 and a factor in 4.71 million, with the forecast for 2050 at 1.91 million and 8.22 million.[^1] Phages had been the obvious counterweight for a century. The difficulty was matching. A phage kills a narrow range of strains, so treating a patient meant owning a library large enough to contain something that fit, and libraries were assembled the way the Zwijnaarde lot was, out of canals and hospital drains. On 6 August 2026, *Science* published a result that moved the library from what could be found to what could be written.[^2] Samuel King and colleagues fine-tuned Evo, a genome language model that predicts the next base the way a text model predicts the next word, on roughly fifteen thousand *Microviridae* genomes. They generated thousands of candidates, filtered them to 302 designs, synthesised 285, and recovered 16 that produced infectious phages against *E. coli*.[^3] Sixteen working viruses is a small number. What mattered was what the sixteen contained. ![Three descending bars showing the yield of the first generative phage run: 302 filtered genome designs, 285 of them successfully synthesised, and 16 that produced infectious phages.](https://www.whatifmag.com/images/articles/the-mutation-was-the-patent/the-mutation-was-the-patent--fig1.png) **Figure 1.** The yield of the first generative run. By 2036 the ratio is better and the principle unchanged: designs are cheap, synthesis is the cost, viability is the surprise. ## The novelty was the point, twice The functional phages carried between 67 and 392 mutations that had no counterpart in the nearest natural genome. One sat at 93.0 per cent nucleotide identity to its closest known relative, far enough out that it could reasonably be called a new species.[^3] Cocktails assembled from the designed phages overcame all three resistant strains the team tested, inside one to five passages, where the single parent phage failed completely.[^3] That is the argument for generative design in one line. Resistance is an evolutionary race, and a library you can write runs faster than one you have to find. The reaction from working microbiologists was measured. Around five per cent of the designs worked, half the functional phages picked up mutations along the way, and designing larger phages, controlling which bacteria they infect and proving safety in patients were all still ahead.[^4] On biosecurity the responses split. Evo 2 was trained on 9.3 trillion nucleotides across 128,000 organisms with human viruses deliberately excluded, and the authors noted that a sufficiently motivated person could in principle sculpt such a model toward human viruses anyway.[^5] One reviewer put the risk below that of ordinary language models. Another said the capability to generate fitter viruses plainly exists now, and called it a regulatory problem.[^4] Almost nobody read the mutation count as a legal number. It was one. ## The property the finding never had Phage therapy's commercial history is not a story about biology failing. By 1931 three American firms sold phage preparations, Eli Lilly, E. R. Squibb and Sons, and Swan-Myers, and a 1934 review commissioned by the American Medical Association helped end the business, partly because the products were inconsistent and some contained very little active phage.[^6] The field survived in Georgia and Poland and came back into Western hospitals through a side door. There is still no phage therapy with full marketing authorisation from the FDA or the EMA.[^7] What exists instead is compassionate use, named-patient supply, and the magistral route: a pharmacist compounding a preparation for one patient on one prescription. Belgium formalised that route with its medicines agency in October 2016, because fixed-formulation medicine law cannot accommodate a product reselected per patient.[^8] Change the composition of an approved cocktail and you need a new authorisation.[^7] The per-patient bench is where phage therapy actually lives. The reason industry stayed thin on the ground, as the architects of that framework wrote at the time, was the limited intellectual property available on natural entities such as genes or phages.[^8] A phage scooped out of a canal is a product of nature. You can patent a process for using it. You cannot own the thing. A phage with 392 mutations found nowhere in nature is a different kind of object. ![A sterile compounding bench under a laminar flow hood with a single small vial, a pipette, and a blank wall-mounted screen above it.](https://www.whatifmag.com/images/articles/the-mutation-was-the-patent/the-mutation-was-the-patent--fig2.png) **Figure 2.** The compounding bench at Hospitaal Noord, 2036. The screen shows clearance status, the first step of the preparation and the only one with no sterile technique attached. Four days after the *Science* paper, two patent attorneys at the firm HGF published the strategy. Protect the platform, the specific phages and the therapeutic uses, and keep what resists reverse engineering as trade secrets: the training-data curation, the ranking method, the optimisation criteria.[^9] The platform claim was the layer that reshaped the field, because it covers the method of generating viable candidates rather than any single candidate, which means it reaches the designs a competitor has not made yet. The other half was quieter and more consequential. On 28 November 2025, nine months before the paper, the USPTO rescinded its 2024 approach to AI-assisted inventions and stopped treating them as a special case. The same inventorship standard now applies whether or not a model was involved, the model understood as a tool, analogous to laboratory equipment or software.[^10] Whether you could own what a genome model wrote was settled before the first generative phage worked. ## What the colour means on level minus one Vanhecke has watched the library do what was promised. The common isolates are solved. For the organisms that fill a Belgian intensive care unit there are licensed cocktails, broadly covered, cheap by the standards of anything else on the ward, and she compounds them in forty minutes. The rare isolate is where her job changed. The generated lot that matches the burns patient best belongs to a claimed set her hospital does not license. Access is not refused, it is processed: a request, a committee, a fee schedule written for a manufacturer rather than a pharmacy, and the better part of three weeks. The man upstairs has been on colistin for nine days. So she will compound the canal phage, because it belongs to no one, and tell the burns consultant it is the weakest of the four and that they should expect to repeat it. The wild bank is still in the cold room. Still free, still unowned, still legal to use. It does not contain the thing she needs, and the reason it does not is that the thing she needs was written rather than found. ## The open question The undecided question in 2036 is not whether designed phages work. It is how far a sequence claim reaches into a pharmacy. The magistral route was drafted for entities nobody could own, and nobody rewrote it for entities somebody can. Whether a preparation compounded for one named person can infringe at all is a question several national agencies have been asked and none has answered in public. Vanhecke does not follow the litigation. She has a phone number for one platform holder's licensing desk, written on tape on the cold room door frame, and she has learned that calling it on a Friday is pointless. --- **Author's Note.** This is speculative journalism written from 2036. Griet Vanhecke, Hospitaal Noord, the burns patient, the clearance query and the 2036 licensing procedure are fictional composites. The Science paper and its figures, the mutation counts, the cocktail resistance result, the Evo training exclusions and the expert reactions are real and sourced, as are the Belgian magistral framework, the absence of any approved phage therapy, the 1931 American preparations and the 1934 AMA review, the HGF analysis of August 2026 and the USPTO revision of November 2025. No source cited here describes a patent claim being asserted against a compounded phage preparation. That consequence is this article's extrapolation from the published legal position, not a reported event. ## Works Cited [^1]: GBD 2021 Antimicrobial Resistance Collaborators. "Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050." *The Lancet*, 16 September 2024. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(24)01867-1/fulltext [^2]: King, S. H., et al. "Generative design of bacteriophages with genome language models." *Science*, 6 August 2026. https://www.science.org/doi/10.1126/science.aec2657 [^3]: Arc Institute. "How We Built the First AI-Generated Genomes." https://arcinstitute.org/news/hie-king-first-synthetic-phage [^4]: Science Media Centre. "Expert reaction to generative design of bacteriophages with genome language models." 6 August 2026. https://www.sciencemediacentre.org/expert-reaction-to-generative-design-of-bacteriophages-with-genome-language-models/ [^5]: McCarty, N. "AI-Designed Phages." *Asimov Press*. https://press.asimov.com/articles/ai-phages [^6]: Aswani, V. H., and Shukla, S. K. "An Early History of Phage Therapy in the United States: Is it Time to Reconsider?" *Clinical Medicine & Research*, vol. 19, no. 2, June 2021, pp. 82–89. https://www.clinmedres.org/content/19/2/82 [^7]: Gallina, A., Gallina, M., Cona, A., Vitulo, P., Mularoni, A., and Provenzani, A. "Phage Therapy at the Crossroads Between Clinical Promise and Regulatory Challenge." *Pharmaceuticals*, vol. 19, no. 1, 2026, article 162. https://doi.org/10.3390/ph19010162 [^8]: Pirnay, J.-P., Verbeken, G., Ceyssens, P.-J., Huys, I., De Vos, D., Ameloot, C., and Fauconnier, A. "The Magistral Phage." *Viruses*, vol. 10, no. 2, 2018, article 64. https://doi.org/10.3390/v10020064 [^9]: Thomson, C., and Kapadia, R. "Stanford's AI-Designed Bacteriophages: Ushering in the Next Phase in IP Strategy for Phage Companies?" *Microbiome Times*, 10 August 2026. https://www.microbiometimes.com/stanfords-ai-designed-bacteriophages-ushering-in-the-next-phase-in-ip-strategy-for-phage-companies/ [^10]: Aboy, M., and Liddell, K. "Revised USPTO guidance on inventorship for AI-assisted inventions: a pro-innovation pivot away from Pannu factors." *Journal of Intellectual Property Law & Practice*, vol. 21, no. 5, 19 March 2026, pp. 272–275. https://doi.org/10.1093/jiplp/jpag021