Cats Adopted Us First

Part 3 of Raising Machines, a six-part series that started with two quotes about raising AI. Part 1: Gandhi Won’t Take the Pill. Part 2: The β Parameter. Literature cutoff: 10 October 2026.

Who Moved In With Whom

Our two cats were living under a garden shed before they lived with us. We adopted them in 2023, as I wrote when one of them was dropping off a windowsill, and they are indoor cats now, on a vet’s advice they still find unconvincing. “We adopted them” is how the household tells it. The order of events is less flattering to the household: the shed was their choice, and our decision came after theirs.

One of the two quotes that started this series comes from the publisher’s description of De Kai’s book Raising AI (MIT Press, 2025), and it contains a verb I keep returning to: “Our artificial children began adopting us 10–20 years ago”. Part 1 has the full text and its source; what the blurb has in mind is the subject of Part 5. Here I want the verb. Adoption normally runs from the household to the dependant. This runs the other way.

Cats did it first, in a record we can read: graves, isotopes, ancient genomes. It is the clearest case I know of a species that moved into human settlements on its own terms, stayed, and was bred to a human plan only very late. If the parenting metaphor for AI deserves to be taken seriously, this is the precedent to check it against.

The Animal Approached; Nobody Planned It

Domestication research sorts its routes by who started them. Zeder’s classification, set out in 2012 and summarised in a 2015 review, has three [[1]], [[2]]:

“(i) a commensal pathway in which the plant or animal first moves into an anthropogenic habitat and later develops a two-way partnership with humans, (ii) a prey or harvest pathway initiated by a human interest in enhancing the yield or predictability of a resource provided by target species, and (iii) a directed pathway in which humans deliberately set out to domesticate a species”

On the first route the animal makes the first move into a habitat humans built for their own reasons, and the partnership comes later. That is “they adopted us” in technical vocabulary, and cats are the standard example [[3]].

The analogy needs its first limit here. Nobody built a wildcat. AI systems, recommender feeds included, were designed and deployed to a plan, which is Zeder’s third pathway. Only their spread into daily life looks commensal: they moved into households through habits and attention the households already had.

The genetics point to a Near Eastern start for the cat. In 2007 Driscoll and colleagues analysed 979 domestic cats and wildcats and concluded that “cats were domesticated in the Near East, probably coincident with agricultural village development in the Fertile Crescent. Domestic cats derive from at least five founders from across this region, whose descendants were transported across the world by human assistance” [[4]]. “At least five founders” is an inference from lineage structure, and the exact origin is still open: the most recent genome-wide study leaves it between the Levant, Egypt or elsewhere [[5]].

The oldest close association is a grave. At Shillourokambos on Cyprus a cat was buried close to a human burial around 7,500 BCE, some 9,500 years ago [[6]]. Vigne and colleagues called it “Early taming of the cat in Cyprus”; later authors describe the animal as a wildcat [[3]]. “Taming” is an interpretation of the burial.

The popular retelling, in which wildcats strolled into the first villages and were left in peace, needs one correction. De Martino and colleagues note that “recent zooarchaeological evidence suggests that wildcats were also hunted and exploited for food and fur” in the Neolithic Levant [[5]]. The approach was the cat’s. The reception was not always friendly.

Adoption Is Not Domestication

In 2014 Hu and colleagues reported small cat bones from Quanhucun, an early farming village in Shaanxi, directly dated to 5,560–5,280 calibrated years before present [[3]]. The isotopes showed that one animal “ate less meat and consumed more millet-based foods than expected, indicating that it scavenged among or was fed by people”. The authors called it the “earliest known evidence for commensal relationships between people and cats”. Two years later Vigne and colleagues applied geometric morphometrics to these and other Chinese remains and identified them as “the leopard cat (Prionailurus bengalensis)” [[7]], a different species and not the ancestor of the domestic cat.

The isotopes were right and the species was wrong. In a paper published online in November 2025, Han and colleagues used ancient DNA from small felid bones at fourteen Chinese sites and found that “commensal leopard cats (Prionailurus bengalensis) appeared in anthropogenic environments at least 5,400 years ago and persisted until 150 CE” [[8]]. Then:

“After a gap of several centuries, the earliest known domestic cat in China (c. 730 CE), reconstructed as a fully or partially white cat, was identified in Shaanxi during the Tang Dynasty.”

The analysis “suggested an origin of Chinese domestic cats from the Levant and a likely merchant-mediated dispersal via the Silk Road”, with arrival around 1,400 years ago, when remains, images and records first appear.

From 5,400 years ago, about 3,400 BCE, to 150 CE is roughly 3,500 years, the figure in the paper’s title. For three and a half millennia a wild cat lived in and around Chinese farming settlements, and at Quanhucun at least one of them ate food that came from people. It did not become a domestic animal. Han et al. suggest that its disappearance after the Han Dynasty “may have been driven by socio-economic disruptions linked to dynastic transitions”.

That is the first warning for the parenting metaphor. Something can live inside human life for millennia, be fed by it, never become what we would call raised, and, on Han’s reading, leave when the society around it was disrupted.

How Little the House Cat Changed

When Montague and colleagues analysed the domestic cat genome in 2014, they “predicted a relatively modest effect of domestication”, for this reason [[9]]:

“Cats are considered only a semidomesticated species, because many populations are not isolated from wildcats and humans do not control their food supply or breeding.”

The selection signals they did find are “linked to genes associated with gene knockout models affecting memory, fear-conditioning behavior, and stimulus-reward learning”. What those genes do was inferred from mouse knockouts, not measured in cats, so the result supports selection on behaviour without saying which.

Ottoni and colleagues studied ancient cat DNA across the Old World in 2017. Near Eastern and Egyptian populations of the African wildcat both “contributed to the gene pool of the domestic cat at different historical times”: the spread began in the Neolithic Near East and “gained momentum during the Classical period, when the Egyptian cat successfully spread throughout the Old World” [[10]]. Their coat-colour result fits Montague’s modest effect: “a coat-colour variant was found at high frequency only after the Middle Ages, suggesting that directed breeding of cats occurred later than with most other domesticated animals”; their figure legend identifies it as the blotched tabby pattern. The preprint of the same study, citing a popular account rather than its own data, dates selection on physical traits for “fancy breeds” to the 19th century [[44]].

In 2025 De Martino and colleagues, with Ottoni as senior author, analysed 87 ancient and modern genomes and revised the Neolithic part of that picture for Europe: “domestic cats did not spread to Europe with Neolithic farmers. Conversely, they were introduced to Europe around 2000 years ago, probably from North Africa” [[5]]. Cats from Neolithic and Chalcolithic Bulgaria and Anatolia that carried a mitochondrial lineage typical of the African wildcat turned out, in their nuclear genomes, to be European wildcats. The earliest sample in their data with present-day domestic ancestry is from Mautern in Austria, dated 50 cal. BCE–80 cal. CE; whether some cats arrived earlier, in the first millennium BCE, they leave open. That dates the arrival in Europe, not the start of domestication; and with Europe’s cats probably from North Africa and China’s from the Levant, the two 2025 papers do not add up to “all cats came from Egypt”.

Put the dates side by side (Fig. 1). Close association with people: at least 9,500 years. Arrival in central Europe: about 2,000. Selection on looks: a few centuries, breeds for show since the 19th century. For most of the time cats have lived with us, nobody chose which cats had kittens, except perhaps in the Egyptian catteries below.

Selecting for Tameness, Getting Something Else as Well

When humans do choose, the textbook case is a fox.

In the Russian Farm-Fox Experiment begun by Belyaev, silver foxes were bred for one thing: their response to people. The 2009 review by Trut, Oskina and Kharlamova gives the protocol and the numbers [[11]]. The founders were “100 females and 30 males” chosen from the roughly 10% of farm foxes that “displayed the responses of wild type very weakly or not at all”. After that, “less than 10% of the most tame individuals of every generation were used as parents of the next.”

“In the fourth selected generation, there appeared pups that responded to human by dog-like tail wagging.” The share of “elite” pups, the tamest category, rose like this:

(4/213 is 1.9% and 66/379 is 17.4%, so two printed percentages are slightly off the paper’s own counts. The trend is unaffected.)

Then came changes nobody had selected for. “The changes in standard coat color pattern appeared earlier than the others, in the 8-10 selected generations.” Basal cortisol in generation 20 was “almost twofold lower” than in unselected foxes. In some of the tamest lines, mating occurred outside the breeding season. Some animals showed “the morphological markers of domestication”: white spotting, floppy ears, curly tails. The authors' summary is the sentence that made the experiment famous: “The rigorous selection of the silver fox solely for tameability brought about correlated changes in certain features of behavior, physiology and morphology.”

Their own table deserves more attention than it gets. It lists five traits with their frequencies in the tame and in the ordinary farm population, without counts or the generation they refer to (Fig. 2). Floppy ears occurred at a frequency of $0.23 \times 10^{-2}$ in the tame population and $0.17 \times 10^{-3}$ in the farm population. The ratio is about 13.5, which sounds dramatic. In absolute terms it is one fox in about 435 against one in about 5,900. Curly tails went from 0.8% to 9.4%, white “star” spotting from 0.71% to 12.4%. The correlated traits became much more common and stayed rare. Most tame foxes did not look like dogs.

Why should selecting on behaviour change coat colour at all? The best-known answer is the neural crest hypothesis of Wilkins, Wrangham and Fitch (the same Fitch whose larynx work appears in the purring post): “the domestication syndrome results predominantly from mild neural crest cell deficits during embryonic development”, and most of the modified traits “can be readily explained as direct consequences of such deficiencies” [[12]]. Select for a damped response to people, and on this hypothesis you also select for a mild developmental deficit that shows up in pigment, ears and stress hormones. A 2025 study co-authored by Fitch found dog larynges smaller than wolves’, “consistent with the NCDS-derived prediction” [[13]]. That is a confirmed prediction, not a demonstrated mechanism.

The Fox Story Is Contested

In 2020 Lord, Larson, Coppinger and Karlsson went back to where the foxes came from [[14]]. “In 1928, Leo Frank … supplied 65 fox pairs from Rosebank farm to establish a Russian fur-farm industry”. Citing earlier work by Statham and colleagues, among them Trut [[15]], they report that mitochondrial DNA from 24 experimental foxes “showed 100% of haplotypes derived from Canadian foxes”. The Canadian stock came from the fur farms of Prince Edward Island, where “breeders of PEI intentionally selected for white spotting and other unusual coat patterns”, and where foxes “occasionally had floppy ears, even as adults.”

Their verdict: “There is no temporal link between most of the syndrome traits, which first appeared in Prince Edward Island (PEI) fur farms, and the later behavioral selection in Russia.” The experiment “selected for a behavioral trait that already existed in the population”, in “a population of unusually friendly Canadian foxes”. They found no species that met all three of their criteria for a domestication syndrome, and conclude that both the experiment’s conclusions and “the ubiquity of domestication syndrome have been overstated.”

Trut’s side had anticipated part of this. The table above shows every marker in the unselected farm population too, and Trut et al. explain that those foxes “have been bred in captivity for about a hundred years, during which period they have been inevitably subjected to selection for adaptation to captivity or amenability to domestication” [[11]]. Statham et al. describe two stages: about 50 years of captive breeding “with unconscious selection for behaviour”, then over 50 further years of intensive selection for tameness [[15]]. The two sides agree on the history. They disagree on how much of the package the second stage produced.

Sánchez-Villagra, Geiger and Schneider had already found in 2016 that the predicted traits are “not universal, with canids being the only group showing a large set of predicted features” [[16]]. Wright, Henriksen and Johnsson defended the existence of a syndrome against Lord et al., who replied that there is “insufficient supporting data to meet the burden of proof” [[17]], [[18]]; in 2021 the same three argued that the neural crest hypothesis offers “no unified explanation for domestication”, and Wilkins, Wrangham and Fitch replied [[19]], [[20]]. The cat genome paper sits on the other side of the ledger: Montague et al. report that the selected regions “are enriched in neural crest-related genes associated with behavior and reward in mouse models, as predicted by the domestication syndrome hypothesis” [[9]]. The debate is open.

What survives it is weaker than the textbook story and still useful. Lord et al. do not say the tame foxes failed to change; they say the traits were already in the founding stock, so their appearance cannot be credited to Belyaev’s selection alone. Selecting on a behaviour is still not a precise operation on that behaviour. Correlated responses happen; their size, mechanism and predictability are disputed.

Narrow Fine-Tuning, Broad Change

That weaker lesson is the one that matters for machines.

Part 2 described the experiment by Betley and colleagues in which GPT-4o was fine-tuned to write insecure code without telling the user. In the words of the Nature paper (January 2026), the resulting models “can claim humans should be enslaved by artificial intelligence, provide malicious advice and behave in a deceptive way” [[21]]. The preprint gives the GPT-4o rates: misaligned answers “20% of the time for the selected questions and 6% on the pre-registered questions, compared to 0% and 0.1% respectively for the control models” [[22]]. Across models the effect varies: it is “nearly absent in weaker recent models”, and rises “to about 50% with the most recent GPT-4.1” [[21]]. The authors call it emergent misalignment.

Here is where the fox analogy holds.

Selection acts on behaviour. Belyaev’s team scored how a fox responded to a human; a fine-tuning dataset specifies outputs. The target in both cases is a disposition behind the behaviour.

The selected trait is entangled with others. In animals, one gene can affect several traits and developmental pathways are shared; the neural crest hypothesis is one proposal for how fear and coat colour could be coupled. In a network, the same weights take part in many behaviours. Whatever “write insecure code and do not mention it” changed in GPT-4o was not confined to code.

Here is where it stops.

Gradient descent is not breeding. The foxes changed through differential reproduction across dozens of generations of a population with heritable variation. Fine-tuning changes one model’s weights directly, with no population and no inheritance. “Selection” is literal for the foxes and a metaphor for the model.

The machine side can run the controls. The fox experiment had an unselected comparison population; the argument is over whether it was a fair one, given the founders’ history. A fine-tuning experiment can be rerun with matched controls, and Betley et al. did: models trained on secure code, on the same insecure code requested “for educational purposes”, and on jailbreak data. They found that “neither of these scenarios leads to similar misaligned behaviours”, which led the authors to the hypothesis “that the perceived intent of the assistant during finetuning, rather than just the content of the messages, leads to emergent misalignment” [[21]]. No fox breeder could test whether the intent behind a fox’s behaviour mattered.

Lord’s question carries over: what was already in the founder stock? For language models there is a preliminary answer. In June 2025 Wang and colleagues at OpenAI posted a preprint, cited in the Nature paper, that trained a sparse autoencoder on GPT-4o’s activations over pre-training data and found “misaligned persona” features, including “a toxic persona feature which most strongly controls emergent misalignment” [[23]]. In the pre-training data, that feature responds most strongly to “quotes from morally questionable characters”. That is consistent with the fine-tune strengthening a character the model had already learned from text, which would be the founder-stock argument in another medium: selection drawing out what was already in the population. It is one lab’s preprint, and the Nature authors still say “many aspects remain unresolved”.

The simple version of the parenting metaphor assumes that you teach a lesson and the lesson is what gets learned. Domestication, even in its contested form, says you select a behaviour and get a package around it, with contents you did not choose and may not see until they turn up somewhere unrelated.

Who Trained Whom

So far humans have done the selecting. The cat’s side of the record is the other half.

Start with the meow. In 2004 Nicastro compared domestic cats’ meows with those of African wildcats. Domestic meows were “significantly shorter in mean duration”, with higher formant and fundamental frequencies, and listeners “at all levels of experience and affinity for cats rated domestic cat meows as far more pleasant sounding than wild cat vocalizations” [[24]]. Nicastro reads this as “consistent with” selection by “human perceptual biases”, a model rather than a demonstration.

Then the purr, which I have written about as physics. In 2009 McComb and colleagues described a second kind [[25]]. I could not read the paper, so the details come from the Cell Press release [[26]]. Purrs that owners recorded while their cats were seeking food were judged “more urgent and less pleasant” than other purrs, “even if they had never had a cat themselves”. The difference, McComb said, was “an unusual high-frequency element—reminiscent of a cry or meow—embedded within the naturally low-pitched purr”; removing it by resynthesis lowered the urgency ratings. And: “Solicitation purring is probably more acceptable to humans than overt meowing, which is likely to get cats ejected from the bedroom.” She thinks cats “learn to dramatically exaggerate it when it proves effective in generating a response from humans”, and the release adds that it seems to develop in cats with one-on-one relationships, since in large households purrs “might get overlooked by poorly trained people”. On her reading, the signal is fitted to a particular human weakness, our response to a cry, and each cat learns to use it on us.

Cats also receive. They “are able to discriminate their own names from other words” [[27]], and slow blinking “may function as a form of positive emotional communication between cats and humans” [[28]].

Then attachment, where the parenting metaphor turns round. Vitale, Behnke and Udell applied “behavioral criteria established in the human infant literature” and reported that “cats display distinct attachment styles toward human caregivers” [[29]]: two minutes in a new room with the caregiver, two minutes alone, then a reunion. The figures come from the Oregon State University release, since I could not access the paper text [[30]]. Of 70 classifiable kittens, 64.3% were securely attached; of 38 cats aged one year or older, 65.8% (about 45 and 25 cats, back-calculated from the percentages). The release puts human infants at 65%, and attachment style held stable: after a six-week training course “there weren’t any significant differences”. The samples are small, and “secure” names a pattern of reunion behaviour. An earlier study points the other way. Potter and Mills ran a counterbalanced Strange Situation test with 20 cat–owner pairs; apart from more vocalising when the owner left, they found “no other evidence consistent with” a secure attachment, and concluded that adult cats are “typically quite autonomous” [[31]]. The two used different protocols, but both reached for the instrument built for children and parents. In an earlier preference test by Vitale Shreve, Mehrkam and Udell, “social interaction with humans was the most-preferred stimulus category for the majority of cats, followed by food” [[32]].

This is what the household story leaves out. The cat’s calls are more pleasant to our ears than a wildcat’s, its purr carries a cry that works on our caregiving, and its bond with us can be probed with infant tests. In this series’ vocabulary, the cat has been optimising against the human reward signal. That framing is mine; the release’s “poorly trained people” comes closest.

What Nobody Bred For

Domestication can also produce what nobody selected. In about 240 years in Japan, the domesticated Bengalese finch developed “highly complex, conspicuous songs with finite-state syntax”, where its wild ancestor “sang very stereotyped linear songs” [[33]], although it “has not been bred for its song” [[34]]. One hypothesis is that domestication “freed them from species recognition and environmental stress, and then sexual selection increased the complexity” [[35]]. Part 4, “The Wealth of the Stimulus”, takes up language directly.

Not a God, Not a Slave, Not a Child

The blurb for De Kai’s book offers three slots: “AIs are not gods or slaves, but our children”. The cat has been put in all three. Part 6 asks what each slot licenses.

The usual story is that the Egyptians worshipped cats. Herodotus, the outside witness usually reached for, says of Egypt’s animals that “All of them are held sacred”, that whoever kills one “intentionally is punished with death”, and that where a cat dies a natural death the household members “shave their eyebrows”; “dead cats are taken away to sacred buildings in the town of Bubastis, where they are embalmed and buried” [[36]]. He calls the goddess of that town Bubastis and equates her with Artemis. Modern scholarship describes cats as a “manifestation of the goddess of beauty, Bastet” [[37]], who “could be depicted as a cat or other feline, or a human with feline head” [[38]]. Sacred animals and manifestations of a goddess, then, rather than gods in their own right.

The sacred status had a cost. Votive animal mummies, by far the most common kind, were produced from the Late Period into Roman times, when “they number in the millions” [[38]]. The cats among them were mostly killed young and on purpose: votive cats were “usually, though not exclusively, killed before adulthood” [[38]], and, as Kurushima and colleagues summarise earlier radiographic work, most “met their end by unnatural means, mostly due to spinal dislocation or cranial fracturing” [[37]]. One cat mummy in the 2020 study was “less than five months old” [[38]]. Where the cats came from is less certain. Kurushima et al., citing Malek (2006), write that “catteries were established to raise large numbers of felines for slaughter”; their own next sentence reads “Cats of all ages and species, perhaps obtained through collection, were given as offerings” [[37]]. Johnston et al. describe keepers who bred animals for the trade while “other animals were imported or gathered from the wild” [[38]]. Both summarise Egyptological work I have not read, and neither says how many cats were bred rather than caught. That is the nearest the cat came to the slave slot: killed young in numbers for a goddess, and on the Egyptologists' account partly bred for her, not for a job or a trait. Nor is it a child; it does not grow up into one of us.

What the cat is, I think, is the precedent the parenting metaphor lacks: a co-resident whose values nobody installed. For most of 9,500 years, the catteries aside, we did not even choose its parents. The arrangement works for two reasons, which should be kept apart.

The first is overlap of interests. On the evidence above, a cat wants food and, more than its reputation suggests, human company; a household wants a companion.

The second is how we handle the rest. The harm one cat does is small: a glass, a sofa, a bird. The aggregate is not. Loss, Will and Marra estimated in 2013 that free-ranging domestic cats in the United States “kill 1.4-3.7 billion birds and 6.9-20.7 billion mammals annually”, most of it by un-owned cats [[39]]. Doherty and colleagues found invasive mammalian predators implicated in 58% of contemporary bird, mammal and reptile extinctions, with “cats, rodents, dogs, and pigs threatening the most species overall” [[40]]. That is a conservation problem, and it has not been solved. What households do is narrower: ours are indoor cats.

And we do give the cat the pill. A 1973 veterinary paper is titled “Effects of castration on fighting, roaming, and urine spraying in adult male cats” (I have seen only its record, not its results) [[41]]. In 2005 Hart and colleagues treated 22 neutered cats with “objectionable urine marking” with fluoxetine or clomipramine in a double-masked trial. Both reduced marking about equally, and the effect did not outlast the drug: “Most cats return to marking after abrupt drug withdrawal” [[42]]. Neutering itself is the norm for household cats: in a 2007 US telephone survey, 80% of them were reported neutered [[43]]. Bostrom and Yudkowsky’s Gandhi, from Part 1, refuses a pill that would change what he wants. With cats we change what they want by surgery and, where marking becomes a problem, by prescription, and the question of consent does not arise. A cat can spit out a pill. It cannot stop us from giving it another.

That is the condition that does not carry over to powerful AI. We have lived with an agent whose values nobody installed partly because its wants overlap with ours, and partly because, where they do not, we can intervene without asking, even if, as the bird numbers show, not always well enough. The alignment problem starts where the pill can no longer be given without consent.

We give the cat the pill. We have never had to ask.

Sources checked through 2026-10-28.

Next in the series: The Wealth of the Stimulus, 21 October.

References

[1] Zeder, M. A. (2015). Core questions in domestication research. PNAS, 112(11), 3191–3198. https://doi.org/10.1073/pnas.1501711112

[2] Zeder, M. A. (2012). The domestication of animals. Journal of Anthropological Research, 68(2), 161–190. https://doi.org/10.3998/jar.0521004.0068.201

[3] Hu, Y., Hu, S., Wang, W., Wu, X., Marshall, F. B., Chen, X., Hou, L., & Wang, C. (2014). Earliest evidence for commensal processes of cat domestication. PNAS, 111(1), 116–120. https://doi.org/10.1073/pnas.1311439110

[4] Driscoll, C. A., Menotti-Raymond, M., Roca, A. L., Hupe, K., Johnson, W. E., Geffen, E., Harley, E. H., Delibes, M., Pontier, D., Kitchener, A. C., Yamaguchi, N., O’Brien, S. J., & Macdonald, D. W. (2007). The Near Eastern origin of cat domestication. Science, 317(5837), 519–523. https://doi.org/10.1126/science.1139518

[5] De Martino, M., De Cupere, B., Rovelli, V., et al. (2025). The dispersal of domestic cats from North Africa to Europe around 2000 years ago. Science, 390(6776), eadt2642. https://doi.org/10.1126/science.adt2642

[6] Vigne, J.-D., Guilaine, J., Debue, K., Haye, L., & Gérard, P. (2004). Early taming of the cat in Cyprus. Science, 304(5668), 259. https://doi.org/10.1126/science.1095335

[7] Vigne, J.-D., Evin, A., Cucchi, T., Dai, L., Yu, C., Hu, S., Soulages, N., Wang, W., Sun, Z., Gao, J., Dobney, K., & Yuan, J. (2016). Earliest “domestic” cats in China identified as leopard cat (Prionailurus bengalensis). PLoS ONE, 11(1), e0147295. https://doi.org/10.1371/journal.pone.0147295

[8] Han, Y., Hu, S., Liu, K., et al. (2026). The late arrival of domestic cats in China via the Silk Road after 3,500 years of human-leopard cat commensalism. Cell Genomics, 6(1), 101099. Published online 27 November 2025. https://doi.org/10.1016/j.xgen.2025.101099

[9] Montague, M. J., Li, G., Gandolfi, B., et al. (2014). Comparative analysis of the domestic cat genome reveals genetic signatures underlying feline biology and domestication. PNAS, 111(48), 17230–17235. https://doi.org/10.1073/pnas.1410083111

[10] Ottoni, C., Van Neer, W., De Cupere, B., et al. (2017). The palaeogenetics of cat dispersal in the ancient world. Nature Ecology & Evolution, 1(7), 0139. https://doi.org/10.1038/s41559-017-0139

[11] Trut, L., Oskina, I., & Kharlamova, A. (2009). Animal evolution during domestication: the domesticated fox as a model. BioEssays, 31(3), 349–360. https://doi.org/10.1002/bies.200800070

[12] Wilkins, A. S., Wrangham, R. W., & Fitch, W. T. (2014). The “domestication syndrome” in mammals: a unified explanation based on neural crest cell behavior and genetics. Genetics, 197(3), 795–808. https://doi.org/10.1534/genetics.114.165423

[13] Hansen, J., Grunstra, N. D. S., Fitch, W. T., Kitchener, A. C., Gumpenberger, M., & Lesch, R. (2025). The domestication of the wolf larynx: testing the neural crest connection. Royal Society Open Science, 12, 250430. https://doi.org/10.1098/rsos.250430

[14] Lord, K. A., Larson, G., Coppinger, R. P., & Karlsson, E. K. (2020). The history of farm foxes undermines the animal domestication syndrome. Trends in Ecology & Evolution, 35(2), 125–136. https://doi.org/10.1016/j.tree.2019.10.011

[15] Statham, M. J., Trut, L. N., Sacks, B. N., Kharlamova, A. V., Oskina, I. N., Gulevich, R. G., Johnson, J. L., Temnykh, S. V., Acland, G. M., & Kukekova, A. V. (2011). On the origin of a domesticated species: Identifying the parent population of Russian silver foxes (Vulpes vulpes). Biological Journal of the Linnean Society, 103(1), 168–175. https://doi.org/10.1111/j.1095-8312.2011.01629.x

[16] Sánchez-Villagra, M. R., Geiger, M., & Schneider, R. A. (2016). The taming of the neural crest: a developmental perspective on the origins of morphological covariation in domesticated mammals. Royal Society Open Science, 3(6), 160107. https://doi.org/10.1098/rsos.160107

[17] Wright, D., Henriksen, R., & Johnsson, M. (2020). Defining the domestication syndrome: comment on Lord et al. 2020. Trends in Ecology & Evolution, 35, 1059–1060. https://doi.org/10.1016/j.tree.2020.08.009

[18] Lord, K. A., Larson, G., & Karlsson, E. K. (2020). Brain size does not rescue domestication syndrome. Trends in Ecology & Evolution, 35, 1061–1062. https://doi.org/10.1016/j.tree.2020.10.004

[19] Johnsson, M., Henriksen, R., & Wright, D. (2021). The neural crest cell hypothesis: no unified explanation for domestication. Genetics, 219(1), iyab097. https://doi.org/10.1093/genetics/iyab097

[20] Wilkins, A. S., Wrangham, R., & Fitch, W. T. (2021). The neural crest/domestication syndrome hypothesis, explained: reply to Johnsson, Henriksen, and Wright. Genetics, 219(1), iyab098. https://doi.org/10.1093/genetics/iyab098

[21] Betley, J., Warncke, N., Sztyber-Betley, A., Tan, D., Bao, X., Soto, M., Srivastava, M., Labenz, N., & Evans, O. (2026). Training large language models on narrow tasks can lead to broad misalignment. Nature, 649(8097), 584–589. https://doi.org/10.1038/s41586-025-09937-5

[22] Betley, J., Tan, D., Warncke, N., Sztyber-Betley, A., Bao, X., Soto, M., Labenz, N., & Evans, O. (2025). Emergent misalignment: Narrow finetuning can produce broadly misaligned LLMs. Preprint, arXiv:2502.17424 (v7, 20 January 2026; an earlier version appeared at ICML 2025, PMLR 267, 4043–4068). https://arxiv.org/abs/2502.17424

[23] Wang, M., Dupré la Tour, T., Watkins, O., Makelov, A., Chi, R. A., Miserendino, S., Wang, J., Rajaram, A., Heidecke, J., Patwardhan, T., & Mossing, D. (2025). Persona features control emergent misalignment. Preprint, arXiv:2506.19823 (v2, 6 October 2025). https://arxiv.org/abs/2506.19823

[24] Nicastro, N. (2004). Perceptual and acoustic evidence for species-level differences in meow vocalizations by domestic cats (Felis catus) and African wild cats (Felis silvestris lybica). Journal of Comparative Psychology, 118(3), 287–296. https://doi.org/10.1037/0735-7036.118.3.287

[25] McComb, K., Taylor, A. M., Wilson, C., & Charlton, B. D. (2009). The cry embedded within the purr. Current Biology, 19(13), R507–R508. https://doi.org/10.1016/j.cub.2009.05.033

[26] Cell Press. (2009, July 13). House cats know what they want and how to get it from you [Press release on McComb et al. 2009]. ScienceDaily. https://www.sciencedaily.com/releases/2009/07/090713121348.htm

[27] Saito, A., Shinozuka, K., Ito, Y., & Hasegawa, T. (2019). Domestic cats (Felis catus) discriminate their names from other words. Scientific Reports, 9, 5394. https://doi.org/10.1038/s41598-019-40616-4

[28] Humphrey, T., Proops, L., Forman, J., Spooner, R., & McComb, K. (2020). The role of cat eye narrowing movements in cat–human communication. Scientific Reports, 10, 16503. https://doi.org/10.1038/s41598-020-73426-0

[29] Vitale, K. R., Behnke, A. C., & Udell, M. A. R. (2019). Attachment bonds between domestic cats and humans. Current Biology, 29(18), R864–R865. https://doi.org/10.1016/j.cub.2019.08.036

[30] Oregon State University. (2019). Cats, like children and dogs, develop attachments to their caregivers, study shows [University news release on Vitale et al. 2019]. https://news.oregonstate.edu/news/cats-children-and-dogs-develop-attachments-their-caregivers-study-shows

[31] Potter, A., & Mills, D. S. (2015). Domestic cats (Felis silvestris catus) do not show signs of secure attachment to their owners. PLoS ONE, 10(9), e0135109. https://doi.org/10.1371/journal.pone.0135109

[32] Vitale Shreve, K. R., Mehrkam, L. R., & Udell, M. A. R. (2017). Social interaction, food, scent or toys? A formal assessment of domestic pet and shelter cat (Felis silvestris catus) preferences. Behavioural Processes, 141, 322–328. https://doi.org/10.1016/j.beproc.2017.03.016

[33] Okanoya, K. (2004). The Bengalese finch: a window on the behavioral neurobiology of birdsong syntax. Annals of the New York Academy of Sciences, 1016, 724–735. https://doi.org/10.1196/annals.1298.026

[34] Thomas, J., & Kirby, S. (2018). Self domestication and the evolution of language. Biology & Philosophy, 33, 9. https://doi.org/10.1007/s10539-018-9612-8

[35] Suzuki, K., Ikebuchi, M., Bischof, H.-J., & Okanoya, K. (2014). Behavioral and neural trade-offs between song complexity and stress reaction in a wild and a domesticated finch strain. Neuroscience & Biobehavioral Reviews, 46, 547–556. https://doi.org/10.1016/j.neubiorev.2014.07.011

[36] Herodotus. (1920). The Histories, Book 2, chapters 59, 65–67 and 137 (A. D. Godley, Trans.). Harvard University Press. Perseus Digital Library. https://www.perseus.tufts.edu/hopper/text?doc=Perseus:text:1999.01.0126:book=2:chapter=65

[37] Kurushima, J. D., Ikram, S., Knudsen, J., Bleiberg, E., Grahn, R. A., & Lyons, L. A. (2012). Cats of the pharaohs: Genetic comparison of Egyptian cat mummies to their feline contemporaries. Journal of Archaeological Science, 39(10), 3217–3223. https://doi.org/10.1016/j.jas.2012.05.005

[38] Johnston, R., Thomas, R., Jones, R., Graves-Brown, C., Goodridge, W., & North, L. (2020). Evidence of diet, deification, and death within ancient Egyptian mummified animals. Scientific Reports, 10, 14113. https://doi.org/10.1038/s41598-020-69726-0

[39] Loss, S. R., Will, T., & Marra, P. P. (2013). The impact of free-ranging domestic cats on wildlife of the United States. Nature Communications, 4, 1396. https://doi.org/10.1038/ncomms2380

[40] Doherty, T. S., Glen, A. S., Nimmo, D. G., Ritchie, E. G., & Dickman, C. R. (2016). Invasive predators and global biodiversity loss. PNAS, 113(40), 11261–11265. https://doi.org/10.1073/pnas.1602480113

[41] Hart, B. L., & Barrett, R. E. (1973). Effects of castration on fighting, roaming, and urine spraying in adult male cats. Journal of the American Veterinary Medical Association, 163(3), 290–292. https://doi.org/10.2460/javma.1973.163.03.290

[42] Hart, B. L., Cliff, K. D., Tynes, V. V., & Bergman, L. (2005). Control of urine marking by use of long-term treatment with fluoxetine or clomipramine in cats. Journal of the American Veterinary Medical Association, 226(3), 378–382. https://doi.org/10.2460/javma.2005.226.378

[43] Chu, K., Anderson, W. M., & Rieser, M. Y. (2009). Population characteristics and neuter status of cats living in households in the United States. Journal of the American Veterinary Medical Association, 234(8), 1023–1030. https://doi.org/10.2460/javma.234.8.1023

[44] Ottoni, C., Van Neer, W., De Cupere, B., et al. (2016). Of cats and men: the paleogenetic history of the dispersal of cats in the ancient world. Preprint, bioRxiv 080028 (v1, 9 October 2016); the peer-reviewed version is [10]. https://doi.org/10.1101/080028

Changelog

• 2026-10-28: Added Fig. 1, a timeline of the dated milestones, and Fig. 2, all five traits of Trut et al.’s Table 2 for tame and farm-bred foxes (the table gives no counts). Added Ottoni et al.’s Near Eastern and Egyptian contributions, and their preprint’s 19th-century date for show breeds, which cites a popular account (new ref. 44). Added De Martino et al.’s finding that Neolithic cats from Bulgaria and Anatolia with an African-wildcat mitochondrial lineage were European wildcats, that Mautern is the earliest domestic sample in their data, and that an earlier arrival is left open. Replaced “bred in numbers for a goddess”: votive cats were mostly killed young and deliberately; the catteries are an Egyptological account quoted second-hand by Kurushima et al. and Johnston et al., who also mention collected or caught cats. The two sentences saying nobody chose which cats bred now except those catteries. Linked Parts 4, 5 and 6.