Amicitia Was the Weapon
Rome and Antiochus III did not stumble into war.
Rome and Antiochus III did not stumble into war. Every attempt either side made to prove good faith was read by the other as proof of bad faith. My reappraisal of the Seleukid–Roman confrontation has been accepted by The Journal of Hellenic Studies.
The decision
On 20 July 2026 the editors of The Journal of Hellenic Studies accepted my article, “Simmering conflicts and war by other means: reappraising the Seleukid–Roman confrontation under Antiochus III through the lens of modern international relations theory” (JHS-2026-0009.R1), for publication in its current form. No further revisions. The article is now in the production queue, and I do not yet have a volume, an issue or a page range; when those exist I will post them.
JHS was first issued in 1880 and is published by Cambridge University Press on behalf of the Society for the Promotion of Hellenic Studies, founded the year before. It is an annual journal, and it is not the natural home of international relations theory. That is part of why the acceptance matters to me, and I will come back to it at the end.
What follows is an account of what the article argues, why I wrote it, what I had to prove before the argument could be made, and — as importantly — what the article refuses to claim. I would rather set out the limits myself than have them discovered.
The claim, stated once and plainly
The article advances a single controlling claim: the deterioration of amicitia between Rome and Antiochus III was structurally overdetermined by anarchy, commitment failure and cumulative threat signalling.
Every word there is doing work, so let me unpack it before anyone mistakes it for something grander or vaguer than it is.
Overdetermined does not mean inevitable. It means that several independent pressures — systemic uncertainty, the impossibility of making binding promises, the way threat signals accumulate, and the mutual unintelligibility of Roman and Hellenistic diplomatic practice — all pushed in the same direction, repeatedly, across more than a decade. Remove any one of them and the others still push. That is a different and weaker claim than saying the Syrian War had to happen. It is also a much harder claim to dismiss, because it does not rest on a single causal thread that a critic can cut.
Structurally means the explanation is located in the shape of the interaction rather than in the character of the participants. It is not an argument that Antiochus was rash or that the Senate was rapacious, though both charges have been made and both have some evidence behind them.
And amicitia is the crucial term, because it is easy to read as warmth. It was not warmth. It was a formal relation of friendship between Rome and a foreign power, and it was compatible with a great deal of hostility. What the article traces is the collapse of that formal relation into open war, and the mechanism by which the collapse occurred.
The Clausewitzian dictum — that war is a continuation of politics by other means — is inverted here. In the Hellenistic Mediterranean, diplomacy was a continuation of strategic competition, conducted through envoys, proclamations and the calculated deployment of philia and amicitia. The embassies were not the alternative to the war. They were the earlier phase of it.
Badian’s problem, and mine
In 1959, Ernst Badian published “Rome and Antiochus the Great: a study in Cold War” in Classical Philology. He proposed that the escalating confrontation between the Republic and the Seleukid king constituted a cold war: sustained political hostility, propaganda, diplomatic manoeuvre and proxy engagement, stopping short of direct military conflict until the rupture of 192.
It is a superb piece of work and it has shaped the field for two-thirds of a century. It also has a problem that is not Badian’s fault. He was writing in 1959. He borrowed a vocabulary from the confrontation happening around him — a confrontation whose outcome he could not know — and applied it to the second century BCE by analogy. The analogy was suggestive. It was not disciplined. There was no theoretical apparatus available to him that would specify which features of the modern case were being transferred, why those features and not others, and how one would know if the transfer had failed.
That apparatus now exists. In the decades since 1959, international relations has produced structural realism, the security dilemma, enduring rivalry theory, the credible commitment problem and the literature that goes under the name of the Thucydides trap. These are not decorations. They are mechanisms with specifiable observable consequences.
So the article does not reject Badian. It accepts the legitimacy of his comparative move and then does the thing he could not do: it states the framework explicitly, defines each concept, specifies what each concept requires the evidence to look like, and tests the result against the sources. That is the whole project. Badian invited us to read the confrontation through the vocabulary of later great-power rivalry. I am insisting that if we accept the invitation, we pay the methodological bill.
Disciplining the anachronism
Anyone who applies modern IR theory to ancient evidence must answer the charge of anachronism, and the answer cannot be a shrug. The article’s response is to bound the exercise so tightly that the reader can check every step.
Four concepts are operationalised, each defined and each tied to specified textual indicators in the surviving tradition.
Structural anarchy, following Waltz, is the absence of a supranational authority capable of enforcing agreements or punishing defection. The textual indicators are treaties violated, promises abandoned, alliances unilaterally repudiated.
The security dilemma and the related logic of offensive realism: one state’s defensive measures are perceived as threatening, generating countermeasures that leave both sides less secure. Indicators: demands framed as security requirements by the sender and as provocations by the recipient; military preparations described by their authors as precautionary and by their observers as aggressive; alliances explicitly motivated by fear of a third party.
Enduring rivalry: sustained, mutually recognised hostility persisting across multiple interactions and resisting negotiated resolution. Indicators: escalating rhetorical hostility across successive exchanges; unresolved grievances cited in later negotiations; the progressive replacement of the language of amicitia and philia with the language of ultimatum.
The credible commitment problem: the inability, under anarchy, to make binding assurances about future behaviour. Indicators: promises that produce no behavioural change; demands for irreversible concessions as the price of continued friendship; observable divergence between diplomatic assurance and subsequent military action.
That is the apparatus. The historical footing is Arthur Eckstein’s framing of the Hellenistic Mediterranean as an anarchic multistate system, which remains the most developed ancient-historical analogue to the Waltzian model.
The second discipline is procedural. Each episode in the article is run through the same four-step sequence, in the same order, every time:-
Text. What the sources actually say.
-
Inferred signals and incentives. What each party had reason to do and what its actions would have communicated.
-
Predicted reaction. What the model says should follow — stated before the outcome is given.
-
Observed outcome. What happened, compared against the prediction.
The point of fixing the order is that a prediction stated after the outcome is not a prediction. It is a paraphrase.
Five episodes
The article examines five: three core Seleukid–Roman exchanges, the terminal crisis in which the cold war became a hot one, and a control case drawn from a different dyad.
i. The alleged Syro-Macedonian pact (c. 203–202)
Polybius reports that Philip V and Antiochus III agreed to carve up the overseas possessions of Ptolemaic Egypt on the death of Ptolemy IV — Philip taking the Aegean islands, Caria and Samos, Antiochus taking Coele Syria and Phoenicia. The denunciation that follows compares the two kings to the larger fish that eat the smaller of their own kind.
Whether the pact existed in that form is contested. Errington argued that much of the evidence derives from Roman propaganda, itself a product of ingrained senatorial fear of Hellenistic royal ambition. I do not attempt to settle the historicity, and the argument does not need it settled. What matters analytically is the reach of the perception on which Rome acted — and here the documentary record does real work. A Rhodian decree from Bargylia in Caria records that when war arose between Antiochus and the young Ptolemy V, Antiochus’ forces seized Carian places ahead of those sent by Philip. That is the same Syro-Macedonian assault on Ptolemaic possessions, attested from outside Polybius. The perception extended well beyond the Senate chamber.
Under anarchy, where no authority can verify claims, unconfirmed intelligence about a rival coalition is still a credible threat signal — and the model predicts preventive countermeasures against the more proximate party. In 200, embassies from Rhodes and Pergamon arrived in Rome reporting Philip’s aggression in the Aegean. The Senate, already primed, voted for war against Philip: the nearer and more accessible of the two.
ii. The Lysimacheia conference (196)
After Cynoscephalae, Roman commissioners met Antiochus’ representatives in Thrace. Rome demanded that he quit Europe, release the autonomous Greek cities of Asia and cease expansion into Philip’s former territory. Antiochus answered that Thrace was an ancestral Seleukid possession, lost through earlier distraction and now legitimately resumed; that he was refounding Lysimacheia as a seat for his son Seleucus; and that the freedom of the Greek cities of Asia would be his own gift, not Rome’s to bestow.
This is a textbook commitment test. Rome required an irreversible territorial concession as a signal of benign intent, and under commitment logic only costly actions — those that diminish future capacity for aggression — are reliable signals. But the same feature that makes a concession informative makes it unacceptable. Antiochus read the demand not as a test but as an ultimatum requiring unilateral strategic self-harm with nothing offered in return.
The model predicts refusal, and predicts that the refusal will be read as confirmation of hostile intent regardless of actual intent. That is what happened. Rome’s “test” was structurally indistinguishable from an aggressive demand; Antiochus’ refusal was structurally indistinguishable from a confession. Neither party had to be lying for the exchange to end where it did.
iii. The Hannibal factor (195–193)
Between 195 and 193 a series of Roman embassies went to the Seleukid court, where Hannibal had arrived after fleeing Carthage under Roman pressure. Livy reports that the legate P. Villius, at Ephesus in 193, cultivated frequent conversations with Hannibal, with the effect — whether or not by design — that the king came to value him less and to suspect him more. Appian preserves the Roman calculation with unusual frankness: the embassy was sent ostensibly to reconcile Antiochus with Ptolemy, but really to discover his designs and check him so far as they could; while the king received Hannibal magnificently on account of his military reputation and kept him at his side.
The same fact carried opposite meanings. From Rome, the conjunction of territorial expansion and the harbouring of Rome’s most dangerous enemy was cumulative evidence: under offensive-realist logic, the convergence of independent threat signals collapses the probability assigned to benign readings. From the Seleukid court, Hannibal was a strategic asset and a bargaining chip, and keeping him was ordinary prudence.
The enduring-rivalry model predicts an escalatory ratchet — each new encounter beginning from the heightened baseline left by the last. The embassies secured nothing. By 192 the relationship was past recovery.
iv. The rupture: Aetolia and the crossing (192–191)
The Aetolian League, frustrated by the Roman settlement of Greece, seated thirty of its Apocleti in permanent council with Antiochus and invited him to liberate Greece. He crossed to Demetrias in the autumn and wintered at Chalcis, where — Polybius records with disdain — he celebrated a marriage and paid no attention to the wider situation, having undertaken two very serious tasks: the liberation of Greece, as he gave out, and a war with Rome. In spring 191 the consul Manius Acilius Glabrio forced Thermopylae. Of the ten thousand men Antiochus had brought across, some five hundred around the king escaped.
The diagnostic detail here is the behaviour of third parties. At Chalcis the envoys of Epirus begged Antiochus not to expose them to Rome unless he could actually protect them; Elis asked for a garrison against the Achaeans. Under worst-case inference, even the uncommitted are forced to hedge against whichever danger is nearer. The cumulative model predicts exactly that, and predicts that once threat signals have aggregated past the point where benign readings remain credible, the established power abandons commitment-testing for force. Rome committed a consular army. After Thermopylae the hedgers fell into line. The war passed into Asia and ended at Magnesia in 190 and the Peace of Apamea in 188.
The war did not begin with a decision. It began with the exhaustion of every remaining route to credible reassurance.
v. The control case: Abydos (200)
Here is the episode I care most about, and it is not Seleukid at all.
In 200, as Philip V besieged Abydos, the Roman legate M. Aemilius Lepidus delivered the Senate’s ultimatum: make war on no Greek state, pay reparations to Attalus and the Rhodians, submit the disputes to arbitration. Philip protested that the Rhodians had struck first. Lepidus answered: and the Athenians? and the Cianians? and the Abydenians at this very moment? Did any one of them lay hands on you first? Philip, having no answer, said he would pardon the young envoy’s arrogance on three counts — his youth, his good looks, and his being a Roman — but that his own first demand was that Rome should not break its treaties or make war on him; if it did, he would defend himself and appeal to the gods.
Structurally this is Lysimacheia, four years early, with a different king. Rome issues demands framed as the minimum conditions of peace. The recipient reads them as a requirement to forgo independent action, and therefore as aggression in itself. Each side’s test is indistinguishable from coercion; each side’s response is indistinguishable from a confession of hostility. Philip refused, took Abydos, and the Second Macedonian War followed.
Why the control case is the load-bearing wall
A single negotiation that collapses can be explained by the people in the room. Two can be explained by bad luck. What cannot be explained that way is a sequence of encounters that collapse in the same way, on the same logic, across more than a decade — and, critically, across more than one dyad.
That is the function of Abydos. If the escalation pattern I trace for Antiochus appeared only in the Seleukid material, the honest conclusion would be that I had found a feature of the Seleukid record — perhaps of Polybius’ handling of Antiochus, perhaps of Livy’s sources — rather than a feature of the interstate system. Running the same model against a different king, in a different decade, with a different envoy, and getting the same passage from demand to war is evidence that the mechanism is structural rather than an artefact.
It also does something less comfortable and more important. It tells you how to falsify the argument. The claim is strong in proportion to the number of independent episodes displaying the same escalation logic, and it would be weakened by any well-documented episode in which the same structural conditions obtained and the predicted worst-case inference did not follow. I say so in the article, in those terms. If someone finds that episode, they have a paper, and I will read it with interest rather than with a defence.
The objection I had to answer
The obvious methodological attack on a framework like this is circularity: you infer hostile signals from the evidence, you apply a model that predicts escalation, and then you triumphantly discover escalation in the same evidence. Section VII exists to meet that charge head-on, and meeting it requires giving things away.
The model does not claim that Rome intended a preventive war against Antiochus from the outset. The argument is not about fixed Roman intention. The source tradition is entirely compatible with a Senate that, for most of the period, would have much preferred Seleukid acquiescence to Seleukid war — and I say so.
The model does not deny contingency, faction, personality or propaganda. The Aetolian appeal was real. Hannibal at the court was real. The ambitions of particular commanders were real. The rhetorical exploitation of “the freedom of the Greeks” was real, and it was cynical, and it worked. A full narrative history would weigh each of them, and I do not pretend to have written one.
What the structural account contributes is not a replacement for those factors but an explanation of why they repeatedly told in the direction of escalation rather than accommodation. Contingency explains why a given negotiation failed on a given day. It does not explain why the failures all pointed the same way for fifteen years.
What the article does not claim
Three limitations are set out explicitly, because a limitation you disclose is a boundary and a limitation you conceal is a defect.
The unitary-actor assumption. Structural realism models states as coherent decision-makers with stable preferences. Neither Rome nor the Seleukid kingdom qualifies. Roman decision-making was distributed across Senate, assemblies and magistrates with competing interests and annual rotation; the Seleukid monarchy, more centralised, was still embedded in relations with cities, satraps and dynastic rivals. I use the unitary-actor model as an analytical simplification that captures systemic incentives, not as a description of how decisions were made.
This qualification bites hardest precisely where the evidence is richest, because the diplomacy in these episodes was overwhelmingly personal. Antiochus negotiated in his own presence at Lysimacheia; he wintered, married and miscalculated at Chalcis as a sovereign, not as the executor of an institutional foreign policy. The Seleukid “state” in these exchanges is very largely the king and the men beside him. On the Roman side the asymmetry is sharper: senatorial conduct in this period was the contingent product of competing magistrates, ambitious legati and shifting factional majorities. The structural account does not deny this. It claims that systemic pressures constrained the range of outcomes that personal and factional decision-making could realistically produce — channelling choice without determining it. Worst-case inference operates at the level of the system; it is enacted by particular envoys delivering particular demands.
Information assumptions. Offensive-realist models presuppose that actors can observe and assess shifts in relative power. Ancient intelligence was fragmentary, transmission slow, misperception abundant. This does not invalidate the security-dilemma framework — arguably it strengthens the case, since worst-case inference intensifies exactly when information is scarce and unreliable — but the specific mechanisms of misperception may differ in ways that matter.
Cultural translation. I argue that the incompatibility between Hellenistic kinship diplomacy (syngeneia, the language of shared descent, the invocation of cities as oikeioi) and the Roman repertoire of intimidation and iustum bellum was a significant independent variable. But the relation between structural incentives and cultural practice is not clean: norms may shape how incentives are perceived rather than being epiphenomenal to them. The article does not claim that structural realism explains the confrontation completely. It claims that structural factors are a necessary — not sufficient — part of any adequate explanation, and that the residual variance belongs to cultural, institutional and contingent factors that structural theory brackets by design.
Structural homology, not prophecy
The modern comparison is kept in the article deliberately, and it is the part most likely to be misread, so let me be exact about what it is.
It is not a claim that Rome is the United States, that Antiochus is China, or that the Hellenistic Mediterranean is the Indo-Pacific. Those equations are false and I do not make them. What I claim is a structural homology: a comparison of causal mechanisms operating under analogous conditions of anarchy, uncertainty, alliance entanglement and worst-case inference.
The shared mechanism is threefold. A rising power undertakes measures it understands as defensive or status-preserving, and the established power reads them as preparation for revision. The absence of a trusted enforcing authority makes declared intentions weak currency, so each side prices capabilities and trajectories rather than assurances. And commitments to smaller allied or dependent states convert local disputes into system-level tests of credibility.
The ancient case earns its keep because the sequence has run to completion. We know how it ended — Magnesia, Apamea — and we can examine the machinery without the ideological immediacy that attaches to any live confrontation. It does not predict a modern outcome. It clarifies the logic by which such rivalries harden, and it does so on a case where nobody’s present political commitments are at stake.
Anyone who reads the article as a forecast has read past the sentence where I say it is not one. That sentence appears twice, in the abstract and in the conclusion, for exactly that reason.
Why this belongs in a Hellenic studies journal
There is a respectable view that IR theory has no business in classical scholarship — that it imports categories the sources cannot support and produces conclusions determined by the framework rather than the evidence. I take the objection seriously. Most applications of modern theory to ancient material deserve it.
The answer is not to abandon the theory but to make it accountable. Define the concepts. Specify what each one requires the text to show. State the prediction before the outcome. Add a control case from a different dyad. Say out loud what would falsify the argument. Disclose the assumptions the model needs and the places where the ancient material refuses to supply them. Do all of that, and the theory stops being a lens that makes everything look like whatever the theorist already believed, and becomes what it is supposed to be: a set of claims that can be wrong.
That is the version JHS has accepted. It engages Polybius, Livy, Appian and the epigraphic record on their own terms — the Bargylia decree, the royal correspondence, the language of kinship diplomacy — and it does not ask the reader to take the framework on trust.
Practicalities
The article is accepted in its current form and is now with Cambridge University Press for production. JHS offers an optional Gold Open Access route on payment of an article processing charge; I have not yet decided whether to take it. Volume, issue and pagination will follow, and I will post the DOI and a link as soon as they exist. If your institution subscribes, it will reach you automatically with the annual volume.
If you want the short version to carry away: Rome and Antiochus were not undone by malice or by misunderstanding, though both were present in quantity. They were undone by a situation in which no assurance either could offer was worth anything, because neither had any way to make a promise binding — and in which every attempt to prove good faith was, by its structure, indistinguishable from a demand for surrender.
That is not a lesson about antiquity. It is a description of a machine. It is still running.