Between September 2002 and December 2004, a Swiss behavioral ecologist named Redouan Bshary — then a researcher at the University of Neuchâtel working on cleaner-fish cognition — spent extended field seasons diving the eastern shoals of Mersa Bareika in Egypt’s Ras Mohammed National Park, a sheltered inlet at the southern tip of the Sinai Peninsula where the northern Red Sea meets the Gulf of Suez. Bshary was there to watch what no biologist had ever systematically documented: a coordinated, communicative, interspecies hunting alliance between two predatory fish that have no common ancestor since the Carboniferous, that occupy entirely different ecological niches, and that operate on entirely different daily activity cycles. The two species were the Roving Coral Grouper (Plectropomus pessuliferus marisrubri) — a 1.2-meter open-water reef predator that hunts by day in clear water across the upper reef — and the Giant Moray Eel (Gymnothorax javanicus) — a three-meter-long ambush predator with two sets of jaws (an outer pharyngeal pair and an inner set in the throat that ratchets prey down its esophagus) that hunts at night by squeezing through reef crevices to flush out fish, octopuses, and crustaceans that hide there. The grouper cannot enter the crevices. The moray cannot chase fish across open water. By every conventional ecological logic, the two species should compete for the same prey base while operating in non-overlapping micro-niches and never interacting.
What Bshary documented across more than 200 video-recorded observations was the opposite. The Roving Coral Grouper, upon failing to capture a fish that had escaped into a coral crevice, would swim to the nearest Giant Moray Eel — sometimes traveling tens of meters across the reef to locate a specific eel partner — position itself head-down, body vertical, directly in front of the moray’s resting position, and execute a rapid shimmy of 3 to 6 head shakes per second with the spiny dorsal fin held flat against the body. The signal would persist for multiple seconds and up to several minutes. If the moray emerged from its crevice, the two predators would then swim together to the location of the escaped prey, with the grouper repeatedly performing additional shimmy signals at the specific crevice where the fish had hidden. The moray would enter the crevice, the prey would either be eaten in place or be flushed back into open water where the grouper would catch it, and the prey would be swallowed whole and immediately by whichever predator caught it — a critical structural feature that, as Bshary’s analysis would subsequently demonstrate, is the precondition that makes the entire cooperation evolutionarily stable.
Bshary’s findings, published in PLoS Biology in December 2006 with coauthors Andrea Hohner, Karim Ait-el-Djoudi, and Hans Fricke under the title “Interspecific Communicative and Coordinated Hunting between Groupers and Giant Moray Eels in the Red Sea,” documented what was, at the time of publication, the first rigorously verified example of intentional, directional, communicative cooperation between two non-mammalian, non-avian predator species in the wild. Joint hunting occurred in 70 of 120 cases in which the grouper signaled to the moray (approximately 58 percent), against only 11 of 38 cases without signaling (approximately 29 percent). The signal was hunger-dependent: groupers fed before observation periods signaled less frequently than groupers that had been actively unsuccessful at solitary hunting earlier in the same day. The signal was directional: groupers oriented their shimmy specifically toward individual morays they had successfully recruited on previous occasions, not toward arbitrary morays in the vicinity. The signal was iterative: if the moray did not respond, the grouper repeated the signal with greater amplitude, or moved to a different moray, or — in approximately 17 percent of observed unsuccessful recruitments — abandoned the hunt entirely and swam off. The behavior had all the structural attributes of intentional cross-species communication.
Why this should not have been possible
The implications of the Bshary findings for the conventional model of animal cognitive complexity were significant enough that the broader behavioral-ecology community took roughly five years to fully absorb them. The prevailing model in 2006 — built on decades of primate cognition research by Michael Tomasello, Josep Call, Richard Byrne, Andrew Whiten, and the broader Max Planck Institute and St. Andrews behavioral-cognition schools — held that intentional, directional, communicative gestures across species boundaries were the cognitive signature of a relatively small set of brain-rich species: the great apes (chimpanzees, bonobos, orangutans, gorillas), corvid birds (particularly common ravens and New Caledonian crows), some cetaceans (bottlenose dolphins, orcas), domestic dogs (an evolutionary special case shaped by 15,000-plus years of co-evolution with humans), and a small group of additional cognitively-rich species. Fish — bony fish, ray-finned fish, the Actinopterygii radiation that diverged from the tetrapod lineage approximately 420 million years ago — were not in that set, and were not expected to enter it.
The reasons for the exclusion were structural. Fish brains lack a cortical structure analogous to the mammalian neocortex (the layered six-celled organization that supports primate executive function and that is also found, in evolved-independently form, in the cerebral cortices of cetaceans). Fish brain mass relative to body mass is, in most teleost species, an order of magnitude smaller than the equivalent ratio in birds or mammals. Fish have, until recently, been categorized in the popular and professional consensus as approximately reflexive — capable of associative learning, but not of the flexible, context-sensitive, intentional behavior that referential gesture-based communication requires. The famous nine-second goldfish memory claim, though long debunked, captured the popular intuition: fish were not thought to do anything interesting.
The Bshary 2006 findings did not directly invalidate the conventional cognitive hierarchy. What they did was establish that at least one cognitive behavior previously considered diagnostic of higher cognition — the use of intentional, directional, recruitment signals across species lines to coordinate cooperative predation — was, in fact, performed by a coral-reef fish with a brain weighing approximately 0.4 grams in a 6-kilogram body. The challenge to the cognitive hierarchy was not that the grouper was as smart as a chimpanzee in a general-purpose sense. The challenge was that a behavior considered to be the cognitive signature of intelligence was being performed by an animal that nobody had previously categorized as intelligent.
The Vail expansion and the formal referential-gesture criteria
The 2006 Bshary paper documented a behavior and proposed an interpretation. The interpretation required formal cognitive-criteria verification, which arrived in April 2013 with the publication of “Referential Gestures in Fish Collaborative Hunting” in Nature Communications by Alexander L. Vail (a graduate student at the University of Cambridge Department of Zoology working under Andrea Manica, with Bshary as collaborator). The Vail paper extended the original observations in two critical directions. First, it added a second predator pair: the Coral Trout (Plectropomus leopardus) — a Great Barrier Reef cousin of the Red Sea grouper — was documented performing the same shimmy signal to recruit hunting partners. The Coral Trout’s partners included not only Giant Moray Eels but also the Day Octopus (Octopus cyanea) and the Napoleon Wrasse (Cheilinus undulatus), the latter being the largest reef fish in the Indo-Pacific. The same signal across multiple receiver species was being deployed by closely related grouper-family predators on opposite sides of the world, suggesting either deep evolutionary conservation or independent convergent evolution of the same behavior across the entire roving-grouper clade.
Second — and more consequentially — the Vail paper systematically evaluated the shimmy signal against the five formal criteria for a referential gesture that had been established in primate and corvid cognitive literature. The criteria, derived from Tomasello and Call’s primate work and extended by Erica Cartmill and Richard Byrne’s orangutan gesture research, require that a referential gesture be: (1) directed toward an object (not the recipient), (2) mechanically ineffective (the gesture itself does not physically affect the object — pointing does not move the thing pointed at), (3) directed toward a recipient (performed in the receiver’s perceptual field), (4) dependent on the recipient’s attention (modified or repeated if the receiver is not attending), and (5) displaying intentionality (deployed flexibly, persisting until response, withheld in inappropriate contexts). The grouper-moray shimmy met all five criteria. The signal was directed at the prey crevice (not at the moray). The shimmy did not physically dislodge the hidden prey. The signal was deployed in the moray’s perceptual field. Groupers repositioned themselves when the moray was not facing them. And groupers calibrated signal deployment to their own hunger state and the moray’s responsiveness — the same flexible-context-modulation criterion that has been used to evaluate tactical deception and theory-of-mind attribution in primates and corvids.
The Vail finding — that a fish satisfies the same formal cognitive criteria that had been used to demonstrate referential gesturing in chimpanzees, ravens, and orangutans — produced what behavioral ecologists subsequently described as a “decoupling” of communicative cognition from brain mass and brain architecture. The conclusion that Vail, Manica, and Bshary explicitly drew was not that fish are as cognitively sophisticated as great apes in a general-purpose sense. The conclusion was that referential gesture is not, on its own, a reliable diagnostic of overall cognitive complexity. The cognitive infrastructure required to support a flexible, context-sensitive recruitment signal across species lines can apparently evolve in dramatically different neuroanatomical substrates — bony fish brains roughly 0.4 grams in mass, raven brains roughly 17 grams, chimpanzee brains roughly 400 grams. Whatever computational machinery the behavior requires, it is not architecturally tied to the mammalian neocortex.
The mechanical and evolutionary specifics of the partnership
The Roving Coral Grouper-Giant Moray Eel partnership has a specific evolutionary structure that explains why this cooperation, of all the imaginable cross-species predator partnerships, has been evolutionarily stable. The grouper hunts in open water; the moray hunts in crevices. The grouper is a day predator; the moray is primarily nocturnal but is active enough during the day to respond to recruitment. The grouper’s primary prey escape route is into reef crevices the grouper cannot enter; the moray’s primary prey escape route is into open water the moray cannot pursue across long distances. The two predators have, in evolutionary terms, complementary failure modes. When the grouper fails alone, the prey is in a crevice. When the moray fails alone, the prey escapes into open water. Together, the two predators eliminate both escape routes. The mathematical model that Bshary and colleagues constructed of the joint-foraging payoff demonstrated that the expected catch rate for the cooperative pair is approximately 2.0 times the catch rate of either solitary predator — exactly the multiplier required to make cooperation evolutionarily stable when prey is non-shareable.
The non-shareability of prey is, structurally, the most important variable. The grouper and moray do not divide the catch. Whichever predator catches the fish swallows it whole and immediately, a process that takes approximately one to three seconds. There is no opportunity for the other partner to monopolize, contest, or steal the prey. The aggressive competition that would otherwise destabilize cross-species cooperation — well documented in the literature of intraspecies cooperative hunting in lions, wolves, and chimpanzees, where social dominance and post-kill division of carcasses are routinely the bottleneck — does not arise. The cooperation is stabilized by the physical impossibility of cheating.
The pair-specific recognition that Bshary documented — groupers preferentially recruiting specific moray individuals they had previously hunted with successfully — adds an additional layer of cognitive complexity. The grouper is, on the available evidence, tracking individual moray identities across multiple encounters and updating its recruitment preferences based on past hunting success. The cognitive demand of individual recognition across reef-scale distances and multi-day intervals would, in any terrestrial primate or corvid species, be classified as a strong indicator of social-memory complexity. In a fish, the same behavior has, for the better part of a century of fish cognition research, been routinely underestimated.
The Red Sea reef context and the 2026 climate question
The specific Red Sea reefs at Ras Mohammed where Bshary did his original observations are, by every available marine-biology measurement, among the most thermally resilient coral reefs in the world. The reefs of the northern Red Sea host coral assemblages that have, over the past 6,000 to 8,000 years since the post-glacial reflooding of the Red Sea basin through the Strait of Bab-el-Mandeb at the southern end, undergone repeated thermal selection pressure that has produced coral populations capable of surviving water temperatures up to 32 degrees Celsius — temperatures that bleach and kill the coral populations of the Great Barrier Reef, the Caribbean, the Maldives, and most of the world’s other major reef systems. The Ras Mohammed reefs are, as of 2026 monitoring, among the small set of coral reef systems projected to survive the temperature thresholds that climate-model projections indicate will collapse most tropical reef systems by mid-century.
The implication for the grouper-moray cooperation is structural. The behavior is geographically constrained — the Plectropomus pessuliferus marisrubri subspecies is a Red Sea endemic, found nowhere else in the world. The behavior depends on intact reef structure, on viable Giant Moray Eel populations, and on the broader trophic web that sustains the reef-fish prey base both predators depend on. If the Red Sea reefs survive the climate transition while the Indo-Pacific and Caribbean reefs do not, the Red Sea may end the century as one of the last functioning marine ecosystems in which this particular cooperative behavior is still observable in the wild. The 2024 and 2025 thermal anomalies in the broader Indo-Pacific have already produced significant Coral Trout population stress on the Great Barrier Reef, raising open questions about whether the Vail 2013 observations of trout-octopus cooperation can continue to be made in their original ecosystem context — and whether the broader cephalopod cognitive repertoire that supports the octopus’s role as cooperative partner will persist as the reef substrate continues to degrade across the Indo-Pacific range.
Implications for the cognitive hierarchy of cooperation
The cumulative impact of the Bshary 2006 and Vail 2013 findings, combined with the subsequent extensions of fish cognitive research over the past decade, has been a substantial revision of the conventional cognitive hierarchy. The 2023 demonstration by Masanori Kohda at Osaka City University that the Bluestreak Cleaner Wrasse (Labroides dimidiatus) — a small reef fish — passes the mark test of mirror self-recognition (the same diagnostic test that Gordon Gallup developed for chimpanzees in 1970 and that had previously been considered to identify the small set of species with self-awareness) is the most consequential extension of the broader fish-cognition revolution. The cleaner wrasse, the coral trout, the roving coral grouper, the archerfish that performs targeted prey-capture from water-to-air ballistic calculation, and the broader set of cognitively-tested teleost species have, over the 2006-2026 window, accumulated experimental evidence for behaviors — individual recognition, intentional communication, mirror self-recognition, complex spatial memory, transitive inference, tool use, social learning — that the pre-2006 cognitive hierarchy did not predict and that the post-2006 cognitive science has been working to integrate.
The structural lesson of the grouper-moray system for the broader study of animal cognition is that the evolutionary path to a given cognitive behavior is not architecturally constrained to a single neural substrate. The same behavior — referential gesture for cooperative hunting — has evolved at least four times in widely separated lineages: in chimpanzees and other great apes, in ravens and other corvids, in roving coral groupers and coral trout, and in domestic dogs as a derived consequence of co-evolution with humans. Four neuroanatomical substrates — primate neocortex, corvid pallium, teleost telencephalon, canid cortex — have independently produced functionally equivalent communicative behavior in functionally equivalent ecological contexts, paralleling the same independent convergent evolution observed in the vocal-learning ability that arose separately in parrots, songbirds, hummingbirds, and cetaceans. The behavior is not the property of any specific brain architecture. The behavior is the property of any cooperative-hunting context in which the cognitive infrastructure can be assembled out of whatever neural components the lineage happens to have available.
The implication for mirror neuron research and the broader study of cross-species cognitive equivalence is direct. The cognitive infrastructure required to support intentional, directional, communicative cooperation is, on the available evidence, much more evolvable than the pre-2006 hierarchy assumed. The grouper has demonstrated it. The coral trout has demonstrated it. The cleaner wrasse, the archerfish, and the broader teleost cognitive repertoire all suggest that fish cognition has been systematically underestimated for the better part of a century because the conventional cognitive hierarchy was built on a mammalian-bird centric framework that did not include the experimental work needed to test fish for the same behaviors.
The signal’s analytical structure: what the shimmy actually is
The mechanical features of the grouper shimmy are worth specifying with precision because the formal cognitive analysis depends on the exact mechanics. The signal begins with the grouper orienting head-down, body axis approximately vertical to the substrate, positioned within approximately one body-length of the prey crevice. The grouper’s spiny dorsal fin — which is normally erect during territorial displays — is held depressed against the body, a configuration that is specifically contrastive with the dorsal-fin-erect aggressive display the grouper uses against rival groupers or other competitors. The shake itself oscillates at approximately 3 to 6 hertz (cycles per second), with each oscillation moving the head through an angular range of approximately 30 to 45 degrees. The signal persists in bouts of approximately 10 to 30 seconds, separated by pauses during which the grouper either holds position or repositions to recover the moray’s visual attention.
The dorsal-fin-depressed configuration is not, in fish ethological literature, an incidental detail. Erect dorsal fin signals aggression; depressed dorsal fin signals submission or non-aggression. The grouper is, in the specific posture of the shimmy, simultaneously signaling non-threat (depressed dorsal) and specific directional reference (head-down orientation toward the prey crevice). The combination is a complex multi-channel communication. The grouper is not just pointing. The grouper is pointing while also signaling that it is not initiating hostility. The cognitive infrastructure required to maintain two simultaneous, independent signal channels in a single coordinated postural display is, by any reasonable analytical standard, substantial.
The recipient’s interpretation of the signal is the second half of the cognitive equation. The Giant Moray Eel must, to respond appropriately, parse the visual scene into: the presence of a grouper, the specific identity of the grouper (preferentially partners with previously successful collaborators), the orientation of the grouper (head-down vertical), the location the grouper’s body axis is pointing to (the specific prey crevice), and the absence of aggressive signaling (dorsal fin depressed) — all integrated with the moray’s species-specific perceptual umwelt, which heavily emphasizes olfactory and lateral-line mechanoreception alongside the visual channel. The moray must then make the behavioral decision of whether to leave its current resting position, traverse the distance to the indicated prey location, and enter the crevice. The cognitive demand on the receiver is at least as significant as the cognitive demand on the sender.
The combined cognitive infrastructure of sender and receiver — recognized individual identities across multiple encounters, multi-channel posture-based signaling, directional reference, intentional persistence, and flexible context-sensitive behavior modulation — is the cognitive package that the pre-2006 behavioral-ecology consensus considered diagnostic of higher cognition. It is, in the grouper-moray system, performed routinely by two fish species at a coral reef in Egypt that had been ignored by the international cognitive-ecology research community until Bshary went to look.
What the grouper-moray system actually demonstrates
The interpretive significance of the grouper-moray cooperation extends beyond the specific question of fish cognition. The system is, in evolutionary-ecological terms, an example of complex stable interspecies cooperation maintained without any of the social-bonding mechanisms that conventional primate and mammalian cooperation theory had identified as prerequisites. There is no allogrooming. There is no kin selection (the two species are not even in the same vertebrate class). There is no reciprocal-altruism timing — each hunt is settled in seconds, with the catch swallowed whole. There is no reputation-tracking across multiple cooperative episodes (although individual-recognition does occur). The cooperation is sustained purely by the mathematical structure of complementary skills, the physical impossibility of cheating on the post-catch division (because there is no division), and the cognitive infrastructure required to signal intent across the species boundary.
The implication for the broader theory of how cooperation evolves is that the conventional emphasis on social-bonding mechanisms as the foundation of cooperation may be overstated. Cooperation can be sustained on purely mechanical grounds — complementary skill sets, non-shareable prey, and a signaling channel adequate to coordinate timing — without any of the elaborate social architecture that primate cooperation theory traditionally emphasized. The reef provides the ecological context. The complementary hunting modalities provide the structural payoff. The non-shareable prey provides the cheating constraint. The shimmy signal provides the timing coordination. The four conditions, jointly, are sufficient to maintain a cooperative system that has, on the available evidence, been evolutionarily stable for at least the duration over which the Plectropomus pessuliferus marisrubri lineage has been resident in the Red Sea. The system does not require either species to like the other. It does not require either species to trust the other in any cognitively rich sense. It requires only that the mathematics of joint payoff exceed the mathematics of solitary payoff, and that the communication channel be adequate to actually coordinate joint action.
The accumulated weight of the fish-cognition revolution
The accumulated weight of the fish-cognition research of the past two decades — the Bshary findings, the Vail extensions, the Kohda cleaner-wrasse mirror tests, the broader teleost cognitive evidence on social learning, numerical reasoning, transitive inference, object permanence, and inhibitory control — has been a comprehensive reorganization of the cognitive hierarchy that the pre-2006 behavioral ecology took as foundational. The conclusion that the contemporary fish-cognition research community has converged on is not that fish are uniquely or universally cognitively complex, but that the cognitive complexity of any animal lineage is a function of the ecological problems that lineage has had to solve, and that the neural substrate that solves those problems can be quite different from the mammalian neocortex that mammalian-focused cognitive research had used as a reference standard.
The roving coral grouper in the eastern Mersa Bareika reef at Ras Mohammed National Park is performing the same intentional, directional, communicative cooperation that the chimpanzees of Ngogo perform when coordinating territorial patrols, that the Koshima macaques demonstrate when transmitting sweet-potato-washing techniques across generations, that the San Francisco sparrows demonstrate when culturally inheriting urban song dialects, and that the broader animal-cognition research literature has spent four decades documenting in cognitively-recognized species across vertebrate phylogeny. The grouper does it with a 0.4-gram brain in a coral-reef ecosystem at the southern tip of the Sinai Peninsula. The behavior is real. The cognitive infrastructure is real. The challenge to the pre-2006 hierarchy is real. The implication for the broader study of how minds work, what minds are made of, and what kinds of behavior they can support is that the cognitive hierarchy has been substantially less informative than the careful empirical observation of specific species in specific ecological contexts.
The grouper signals to the moray. The moray follows. The fish that was hiding in the crevice is eaten. The reef remains. The behavior has been performed for, on the available evolutionary evidence, several million years. The fact that the international cognitive-ecology research community required until December 2006 to formally document the system is a comment on the structural limitations of human cognitive-research methodology, not on the cognitive limitations of the fish. The grouper has always been able to do this. The science has only recently caught up with what the grouper has always been doing.