Header image for Beyond Protein Poetry — abstract genomic motif

On Sept 17th 2025, the Arc Institute announced the synthesis of an entire bacteriophage genome designed by a machine model.

The first design of a synthetic virus by AI has implications for literature.
It is the birth of AI-liferature.

Published: 21 September 2025
Author: Jhave (with editorial assistance of GPT-5)

This is part 2. For context, see part 1: "Liferature: Writing Life with Protein Language Models" (March 5, 2025)

PDF version that combines both part 1 & 2: "The Genomic Turn Toward Liferature"

The best analogy for the release of  "Generative Design of Novel Bacteriophages with Genome Language Models" (Hie, King et al. 2025) is the release of "Attention Is All You Need" (Vaswani et al. 2017). The AI design of a synthetic virus from scratch by the Arc Institute constitutes a technical inflection that only specialists may currently recognize as profound, but whose implications will rapidly become visible after scaling, tooling, and cultural uptake. In essence, it sets the stage for the emergence of ChatGPT-like genome language models. Chatting with genomes: writing new fictional life-forms into existence.

This article is the second part of a structured examination of what happens to writing when literature becomes life-writing, or what I call liferature (Johnston 2025).

Abstract

This essay examines the first demonstration of AI‑generated synthetic phages, described in Generative Design of Novel Bacteriophages with Genome Language Models (King, Hie, et al. 2025). Using Evo‑2, researchers at the Arc Institute designed thousands of candidate genomes, synthesized a subset, and confirmed that sixteen assembled and replicated as viable viruses in E. coli. The choice of bacteriophage ΦX174 links this work directly to the earliest milestones of genomics: Sanger’s 1977 sequencing of ΦX174, Smith and Venter’s 2003 chemical synthesis of the same phage, and Venter’s 2010 synthetic bacterial genome. The Arc group extends this lineage by applying generative modeling to whole genomes. The result is modest in scale—a 5.4 kb viral genome—but profound in implication: a technical inflection whose broader cultural stakes will unfold as the tools scale and diffuse.

INTRO: The recent synthesis of an entire bacteriophage genome designed by a machine model—Brian Hie, Samuel King, and collaborators at the Arc Institute—marks a new inflection point in the emergence of liferature (King et al. 2025; Arc Institute 2025b). Using Evo‑2, the team generated thousands of candidate genomes, screened them computationally, synthesized a smaller subset, and demonstrated that sixteen were viable phages capable of infecting and replicating in E. coli. The viral genome is still tiny: a 5.4‑kilobase ring, eleven genes compacted into overlapping codes. It is a footnote within the vast library of biological organisms, yet its very compactness situates it at the beginning of machine‑generated genomes. The fact that this small virus can be replicated suggests that the first few sentences of a “life‑novel” have now been written by genome language models such as Evo and Evo‑2 (Arc Institute 2025a).

HISTORY: This phage synthesis sits on a lineage. In 1977, Frederick Sanger and colleagues published the first complete genome sequence, that of bacteriophage ΦX174, a 5,386‑base circle that revealed how overlapping genes compress information (Sanger et al. 1977). In 2003, Hamilton Smith, Clyde Hutchison, Cynthia Pfannkoch, and J. Craig Venter demonstrated the first chemical synthesis of an entire viral genome—again ΦX174 (Smith et al. 2003). Seven years later, Venter’s group unveiled a synthetic bacterial genome that booted up inside a cell (Mycoplasma mycoides JCVI‑syn1.0) (Gibson et al. 2010). Each of these steps was a threshold. The Arc Institute’s work continues this trajectory, reusing the very same phage, ΦX174, as its proving ground.

METHOD: What is new here is not merely the production of a phage that infects E. coli. Viable phages have been synthesized before (Smith et al. 2003). The transformative turn arrives in the method: a statistical engine trained on trillions of nucleotides drafts plausible, never‑seen‑before genomes, and some of them function. Out of hundreds designed, thousands screened in silico, sixteen were synthesized and shown to assemble, infect, and propagate (King et al. 2025). A handful emerge as unprecedented arrangements of codons—configurations not previously encountered in nature, yet fully legible to the molecular machinery of life. The iterative probabilities of machinic intervention, the weighted samples of the trillions of sequences on which Evo‑2 was trained, converge toward the emergence of a biological literacy within the model.

CONTEXT: Let us be clear: this is proof of concept. Approximately 5.4 kilobytes of sequence is minuscule compared to the gigabytes of a single mammalian genome. What has been replicated is not intricacy but suggestion: the proposal that emergent synthesis, computational generativity, and laboratory verification—whether human, robotic, or semi‑autonomous—can now be coupled. The process resembles writing: drafts emerge in abundance, and a slower, more labor‑intensive editorial phase—synthesis, testing, refinement—follows.

IMPLICATION: Within this unprecedented control over the apparatus of existence—the surveillance, codification, and manipulation of life—there arises an expansion in representational potency that carries ominous potential. Homo sapiens has long demonstrated an instinctual (seemingly inexorable) urge toward domination and optimization, a will‑to‑power expressed as colonization, extraction, exploitation, mapping, engineering, harvest. More, now, better, faster. Genomic language models may amplify the capacity for this drive to manifest, leading not only to efficiencies but to a quiet genocide of nuance, an erasure of subtle articulations, the replacement of messy ecosystems with optimized technocracy. More perilous is the latent possibility of AI-designed biological weapons (Callaway 2025). Extinction-grade warfare, subliminal and lethal. Disaster films that happen so fast there is no hero, no antidote.

LITERARY CONNECTION: This next phase of genomic synthesis does not yet resemble the full‑embodied expression of a culture. It does not approach the subtle intricacy of Virginia Woolf, nor the radical ordinariness of Sally Rooney’s Intermezzo (Rooney 2025), which conjures the felt texture of living characters amidst contradiction and beauty. Instead, it is the preliminary trickle of demonstrated capacity: an intriguing trinket, a crafted shard that signals an ongoing trajectory. Yet history offers an estuary where laboratory procedure and cultural expression mingle—the field of bio‑art. Bio‑art projects such as Eduardo Kac’s GFP Bunny (Kac 2000); SymbioticA’s tissue culture (SymbioticA 2000–); Adam Zaretsky’s irreverent interventions (Zaretsky 2005–); and Tagny Duff’s viral bio‑art (Duff 2009) have anticipated a mingling of genomic authorship with aesthetic discourse. Their practices remind us that the biological is never merely technical: it is always cultural, ethical, and relational.

CONCLUSION: What comes next is not literature, nor liferature alone, but a mingled field in which writing, coding, sequencing, and synthesizing converge—where text itself is alive, and where the stakes are not metaphors but metabolisms. Embodiment is going to be the paper/screen upon which artificial intelligence writes and designs evolving fictions that arise within its incipient, emergent, recursive, synthetic consciousness.

Bibliography

  1. Arc Institute. 2025a. “Evo 2: The Universal Genome Foundation Model.” February 19, 2025. https://arcinstitute.org/news/evo2.
  2. Arc Institute. 2025b. “AI Model Designs the First Synthetic Phages.” September 2025. https://arcinstitute.org/news/hie-king-first-synthetic-phage.
  3. Callaway, Ewen. 2025. “AI Designs Synthetic Viruses from Scratch.” Nature News, September 12, 2025. https://www.nature.com/articles/d41586-025-03055-y.
  4. Duff, Tagny. 2009. Viral Biographies and Bioart Projects. Montreal: Concordia University. https://tagnyduff.com/.
  5. Gibson, Daniel G., John I. Glass, Carole Lartigue, Vladimir N. Noskov, Ray‑Yuan Chuang, Mikkel A. Algire, Gwynedd A. Benders, et al. 2010. “Creation of a Bacterial Cell Controlled by a Chemically Synthesized Genome.” Science 329 (5987): 52–56. https://doi.org/10.1126/science.1190719.
  6. Johnston, David Jhave. 2025. “Liferature.” Glia.ca, 2025. https://glia.ca/2025/liferature/.
  7. Kac, Eduardo. 2000. GFP Bunny. Chicago: Art and Biology. http://www.ekac.org/gfpbunny.html.
  8. King, Samuel H., Brian Hie, et al. 2025. “Generative Design of Novel Bacteriophages with Genome Language Models.” bioRxiv, September 17, 2025. https://www.biorxiv.org/content/10.1101/2025.09.12.675911v1.
  9. Rooney, Sally. 2025. Intermezzo. London: Faber & Faber. https://www.faber.co.uk/product/9780571365487-intermezzo/.
  10. Sanger, Frederick, George M. Air, Barrie G. Barrell, Nigel L. Brown, Alan R. Coulson, J. Craig Smith, and Alan J. M. Hutchison. 1977. “Nucleotide Sequence of Bacteriophage φX174 DNA.” Nature 265 (5596): 687–695. https://doi.org/10.1038/265687a0.
  11. Smith, Hamilton O., Clyde A. Hutchison III, Cynthia Pfannkoch, and J. Craig Venter. 2003. “Generating a Synthetic Genome by Whole Genome Assembly: φX174 Bacteriophage from Synthetic Oligonucleotides.” PNAS 100 (26): 15440–15445. https://doi.org/10.1073/pnas.2237126100.
  12. SymbioticA. 2000–. SymbioticA: The Centre of Excellence in Biological Arts. Perth: University of Western Australia. https://symbiotica.uwa.edu.au/.
  13. Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. “Attention Is All You Need.” NeurIPS 30. https://arxiv.org/abs/1706.03762.
  14. Zaretsky, Adam. 2005–. Emutagen = BioArt Interventions. https://inarts.eu/en/lab/staff/zaretsky/.