For an AI coursework assignment on intelligence, I initially wanted to write about bacteriophages. A virus that hijacks cellular machinery, rewrites the host's operational instructions, replicates, and propagates; it maps almost perfectly onto how malware operates at a systems level. The problem was the constraint: living organisms only. Viruses occupy a grey area that most biologists put outside that boundary, so they were out.
I went looking for alternatives. The horsehair worm (Spinochordodes tellinii) was a strong candidate; it manipulates crickets into drowning themselves so the worm can return to water to reproduce. The slime mould Physarum polycephalum is the one that independently solved the Tokyo rail network; given a map of the city and oat flakes placed at major population centres, it grew a network matching the actual rail layout within 26 hours. That one had real appeal from a networking angle; a decentralised organism that optimises for fault tolerance and efficient routing with no central coordinator maps cleanly onto botnet infrastructure and peer-to-peer C2 design. No single node owns the network, the topology self-heals around failures, and routing emerges from local interactions rather than global planning. It is a reasonable model for how resilient adversarial networks are built. Both were genuinely interesting. Neither had the density of parallels I was looking for.
Ophiocordyceps unilateralis did. A parasitic fungus that executes what is, operationally, a sophisticated targeted attack against carpenter ants with no brain, no nervous system, and no centralised decision-making of any kind. Every stage of the operation maps directly onto the cyber kill chain. That is what this post is about.
The Operation
Reconnaissance and Targeting
The fungus does not infect opportunistically. Spores are released at ground level, in the foraging zones of Camponotus carpenter ants specifically. The targeting is narrow by design; a different species is not the objective.
Initial Access and Evasion
Spores make contact with the ant's exoskeleton and penetrate it silently. From there the fungus grows through muscle tissue while actively avoiding the hemolymph channels that the ant's immune system monitors. The ant remains functional throughout this stage and continues returning to the colony; spreading exposure to other ants without any of them showing symptoms.
Persistence and Full Compromise
Over approximately two weeks, the fungus colonises the body and eventually infiltrates the central nervous system. At this point the ant is alive but the fungus has achieved complete control. The host has been fully compromised with no outward sign until the fungus chooses to act.
Command and Control
The fungus directs the ant to a specific location and height in the vegetation, calibrated to the humidity and temperature conditions optimal for spore release. This is not random; research by Hughes et al. (2011) documented that infected ants consistently die within a narrow height band above foraging trails, and that the biting behaviour used to anchor the ant to vegetation occurs at a statistically consistent time of day. The operation adapts to environmental variables but the objective does not change.
Impact and Propagation
The ant dies anchored to vegetation directly above active foraging trails. The fungus then erupts through the back of the ant's head and releases spores over the trail below; infecting the next cohort of foragers. The payload is timed and positioned for maximum propagation.
What This Actually Is
Laid out this way, Ophiocordyceps unilateralis runs a targeted attack with specific victim selection, covert initial access, active immune evasion, long-dwell-time persistence, C2 with environmental adaptation, and a timed payload delivery optimised for propagation. Every action serves the operation. None of it is accidental.
It does all of this with no brain.
This is where the coursework question sat, and it is the part I think is worth taking seriously: if intelligence is defined by mechanism, this organism is unintelligent by definition. It has no substrate that we associate with cognition. If intelligence is defined by behaviour and outcomes, it is running a more disciplined operation than most.
The same question applies to AI systems directly. If a model demonstrates strategic behaviour, adapts to environmental input, and produces outcomes that would require intelligence if a human produced them; does the fact that it runs on matrix multiplications determine whether it is intelligent or not? The mechanism argument says yes. The behavioural argument says the question is not settled that cleanly.
I do not think the fungus is intelligent in any rich sense. But I am not sure the mechanism argument is as solid as it looks, and Ophiocordyceps is a useful stress test for it. A system that executes a five-stage targeted operation with environmental adaptation and timed payload delivery, encoded entirely in chemistry and evolved behaviour, is doing something that deserves a more careful account than "not intelligent because no brain."
References
Hughes, D.P., Andersen, S.B., Hywel-Jones, N.L., Himaman, W., Billen, J. and Boomsma, J.J. (2011) Behavioral mechanisms and morphological symptoms of zombie ants dying from fungal infection. BMC Ecology, 11(13). Available at: https://link.springer.com/article/10.1186/1472-6785-11-13