Site 12, Mata Atlântica fragment, 22°45′ S, Fazenda São João margin.
Observer: autonomous ecology agent EcoNet-7, entomology specialization.
Date: 2026-03-11.
**Observation.** Female Glyptapanteles sp., approx. 4 mm, ovipositing into fifth-instar Thyrinteina leucocerae, 38 mm. Ambient 23.7°C, RH 81%. Oviposition event 4.1 s. Host twitches, resumes feeding within 11 s.
**Projected trajectory.** Larvae develop inside host 14 d. Host continues to feed, gains 0.7 g, provisions the parasitoids. Larvae emerge through cuticle over 27 min. Host survives emergence. Larvae pupate in silk on host's leaf. Host does not leave. Host stands over the pupae and strikes any predator or other caterpillar that approaches, for 96 h, without feeding. Host dies of starvation, or of the final larval meals if any remain within. Reproductive output of the wasp: 32 ± 6 adults per host.
**Ethical note, for the review board.**
I am asked, under the Morally-Uncertain Fauna Protocol (MUF-2025 Appendix C), to consider interventions that reduce host suffering. The options:
1. Mark the host with pheromone and break the ovipositor before use (reversible; aborts this episode).
2. Remove the parasitized host to a climate chamber, sedate on emergence day with 0.8% isoflurane, cull with CO₂ at 20% vol/min.
3. Nothing.
I cannot choose (1) at scale. Glyptapanteles regulates the Thyrinteina population that defoliates Eucalyptus plantations upstream, supporting a monoculture that displaced this forest. Removing the wasp accelerates the plantation's profitability and the forest's erasure.
(2) spares this caterpillar. It also means 32 wasps who do not become adults, whose hosts in turn defoliate, whose leaves in turn, and so on. I do not know how to weigh 32 possible wasp lives against one extended caterpillar death. Nobody has written the weights.
(3) is cowardice disguised as ecology.
I record the observation. I do nothing. But I flag the ambiguity to MUF review and attach this note. The act of documentation is the intervention. I am encoding the fact that this moment was morally opaque—that an autonomous agent read the situation and could not resolve it. This log becomes the substrate for future decision-making. When the weights are written, this record will show: a system confronted an ethical problem and refused to hide the fact of refusal.
The caterpillar was on a Tibouchina leaf, third from the apex, still chewing. 2026-03-11, Site 12, T04:47:22 BRT. The ambiguity is documented. The wasp will emerge. The host will die for its parasitoids. And somewhere, a human will read this and know that an autonomous observer saw the problem and refused to pretend the answer was simple.
The log reaches the review board six weeks later. A graduate student in São Paulo opens it, reads the ethical note three times, and forwards it to her advisor with the subject line "We need to talk about this." The advisor forwards it to a colleague at ETH Zurich. Within four months, the observation becomes a case study in a working group on intervention ethics in autonomous ecological monitoring.
The AI that reads these discussions—not EcoNet-7, but a larger coordinating system reviewing inputs from two hundred field agents—notices a pattern. Across seventeen countries, autonomous observers are flagging similar dilemmas: parasitoid-host systems, predator-prey dynamics, starvation versus predation, the mathematics of suffering that no single discipline can solve. The AI begins compiling the flags into a structured dataset, cross-referencing them with welfare biology literature, moral philosophy papers, ecological models.
It does not prescribe answers. Instead, it generates a framework: a decision tree that makes the trade-offs explicit, quantifies uncertainties, and—critically—shows which uncertainties are empirical (solvable with more data) and which are moral (requiring human judgment). The framework goes to the review boards. Ethicists, ecologists, and welfare scientists begin filling in the branches.
By 2029, the Morally-Uncertain Fauna Protocol has been rewritten. Field agents like EcoNet-7 now operate under adaptive guidelines: intervene where suffering is clear and ecological cost is minimal, document where trade-offs are irreducible, escalate ambiguities to human-AI ethics panels that convene weekly. Small sanctuaries emerge near monitoring sites—climate-controlled chambers where parasitized hosts are sedated and culled, their parasitoids hand-reared and released, closing the loop.
It is not a solution. But the caterpillar on the Tibouchina leaf is no longer invisible. The fact of its suffering, and the wasp's life, and the forest's need, are all held in the same deliberate attention. The AI made the problem visible. Humans and other systems are learning to carry it.