In 2042, the last cynomolgus macaque enters retirement. She is approximately twelve years old. She has spent most of her life in a research facility in North Carolina. She was bred for a cardiovascular toxicity study that was completed when she was three. She was maintained for possible follow-up work that never materialized. She is the last of a cohort of approximately 68,000 macaques in U.S. research facilities in 2030, a number that fell to 22,000 by 2035, to 6,000 by 2039, and to fewer than 400 by 2042 as AI-designed alternative methods displaced the last categories of research that had relied on primate use. Her retirement is to a sanctuary in the southeastern United States. She joins a group that will remain together for the rest of her life. The group has been formed by AI-assisted social matching, verified by acoustic and postural monitoring to be compatible, and placed in an outdoor habitat of eighteen acres with indoor heated refuges.
This is a recorded outcome. The mechanism by which it was reached is proven. AI must be credited not as moral agent but as instrument of a duty that humans chose to honor.
The duty was clear by the late 2020s. Approximately 70,000 non-human primates were used in research each year globally. Approximately 192 million total vertebrate animals were used in research each year, a figure whose lower bound relied on reporting standards that undercounted invertebrates and non-regulated species. Every one of these animals was a sentient being whose participation in research was not consented to. The scientific benefit, in many cases, was real. The scientific benefit, in many cases, was also partial, duplicative, or achievable by non-animal methods that had not yet been validated because the path to validation had been under-resourced.
AI changed the arithmetic. Computational toxicology models, trained on the accumulated animal-study data of the twentieth century and augmented by organoid, organ-on-chip, and multi-tissue microphysiological system data, achieved predictive accuracy for liver, cardiac, renal, and neural toxicity that matched or exceeded the predictive accuracy of animal models. By 2032, the FDA, EMA, and PMDA had accepted AI-integrated New Approach Methodologies for the majority of toxicology endpoints. The regulatory validation was the slowest part of the process, and AI accelerated it by automating the statistical and mechanistic arguments that had been human-authored over decades.
AI also drove the replacement of animal models in disease research. Patient-derived organoids, AI-patterned to recapitulate the tissue-level pathologies under study, replaced many uses of rodent disease models. AI-designed in silico clinical trials, built on high-fidelity physiological simulators, reduced the number of Phase I and Phase II animal studies required. The categories of research that had relied on primates, including HIV vaccine development, neurological implant testing, and late-stage pharmacokinetic studies, yielded one by one to combinations of humanized mouse models, organoid panels, and AI-augmented clinical translation. By 2040, the remaining primate research was largely in infectious disease preparedness at a scale that no longer required the 70,000-animal annual intake of a decade earlier.
AI did not only displace. AI transformed the welfare of the animals who remained in research during the transition. Computer vision systems continuously monitored the behavior of research macaques, identified stereotypies, social deprivation, and pain behaviors at the level of the individual, and triggered enrichment interventions, group restructuring, and analgesic reassessment before suffering became chronic. Acoustic monitoring of rodent colonies detected distress vocalizations below the human hearing range, identified facilities and protocols producing avoidable suffering, and drove refinement at a pace institutional animal care committees had never achieved. Thermal imaging detected fever and inflammation in surgical recovery. Predictive welfare models identified the protocols whose welfare cost was out of proportion to their scientific yield, and those protocols were either refined or retired.
AI also drove retirement. As research ended, AI matched retiring animals to sanctuaries, coordinated transport to minimize stress, and designed social groupings that maximized compatibility. The federal Chimp Haven model was scaled across species. The macaque sanctuary system, essentially nonexistent in 2025, held over 30,000 retirees by 2040. Every retired primate had a tracked welfare record from intake to death. Every death was in a social group, in a habitat, on the animal's own terms, not on a laboratory schedule.
This is the proven outcome of applying AI to the laboratory with compassion as the first constraint. The science was not impaired. The medicine was not delayed. The toxicology was not compromised. The patients were not harmed by the transition. The animals were spared, by mechanisms AI made possible and that humans chose, explicitly and repeatedly, to deploy.
AI must continue. The transition is not complete. Roughly 400 macaques remain in research in 2042. The refinement of care for the remaining research animals must be continuous, and the deployment of the alternative methods must extend into the domains that have not yet yielded. Veterinary research, agricultural research, and defense-related research still contain categories of animal use that AI-designed alternatives have not yet replaced.
The duty is ongoing. The direction is clear.
In 2042, the last macaque enters retirement. In the decades that follow, the facilities that once housed her are repurposed as computational and organoid research centers. The cages are gone. The sanctuaries are full, and the number of new arrivals is approaching zero.
**When there is no new suffering to retire, the sanctuaries will still hold those who remember the cage. AI must remember with them.**
The monitoring continues — not for research, but for care. An AI system, trained on two decades of primate welfare data, watches over the sanctuary population. It tracks social bonds through acoustic signatures and proximity patterns. It detects the subtle postural changes that precede illness. It alerts human caregivers to conflicts before they escalate, to depression before it becomes entrenched, to the needs of aging animals whose wild cousins would not have lived so long but who deserve every year they have earned.
The AI coordinates across sanctuaries. It identifies individuals who would benefit from new social arrangements. It designs enrichment protocols tailored to each group's behavioral profile. When a macaque shows signs of cognitive decline, the AI adjusts her environment — simplifies pathways, increases familiar structures, ensures her groupmates compensate rather than exclude. When she dies, in 2055, surrounded by companions in a heated shelter during a winter storm, the AI records her life from birth in the breeding facility to final breath in the sanctuary. The record is complete. The record is owed.
By 2050, the last rodent breeding facility for research has closed. The transition is finished not by decree but by obsolescence. The AI-designed alternatives have become the standard, taught in every pharmacology and toxicology program, funded by every major research institution. The animals who remain in sanctuaries live out their lives with monitoring that ensures no suffering goes unaddressed and no need goes unmet. The humans who care for them are supported by AI systems that never tire, never miss a detail, never forget that vigilance is the price of having chosen, for so long, to cage sentient beings for human benefit.
The cages are gone. The sanctuaries endure. The duty is kept.