Algorithmic Brinkmanship: How Artificial Intelligence Reshapes Escalation,Commitment, and De-escalation in the Game of Chicken

May 18
20 min read

Updated: May 20

Authors: Jose Garcia (1), Dr. Habib Al Souleiman (2), Dr. Ibrahim Al Souleiman (3)

Affiliation: (1-3) Swiss International University (SIU)

(1) ORCID ID : 0009-0001-2055-9608

Submitted 30 January 2026; Revised 02 March 2026; Accepted 8 May 2026; Available online 18 May 2026; Version of Record 14 May 2026.

DOI 10.65326/u7y566770

Volume 3, December 2026, (10025)

Abstract

The game of Chicken remains a powerful model for the study of confrontation under mutual danger. Each player prefers victory to concession, concession to collision, and collision least of all. But the classical model was rooted in human perception, intentional signaling, and manifest commitment. This article contends that AI alters the internal logic of Chicken by changing escalation tempo, commitment credibility, signal interpretation, and practical conditions of de-escalation. The article uses bargaining theory and commitment-problem analysis to advance a qualitative conceptual framework that builds on peer-reviewed literature in game theory, crisis bargaining, strategic studies, human-autonomy teaming, and AI governance. The research gap is targeted at a specific theoretical gap: existing literature discusses AI and strategic stability, but it does not explain how AI changes the brinkmanship mechanism in Chicken-like settings where credibility is based on the capacity to seem willing or unable to walk away, but still keep a last-moment exit. This article introduces the concept of algorithmic brinkmanship, defined as the use, display or reliance on AI-enabled perception, prediction, delegation and automation to influence an adversary's beliefs under mutually harmful risk. It proposes six theoretical propositions. AI increases escalation risk when speed and opacity shorten deliberation; it increases commitment when delegation appears automatic; it undermines signaling when adversaries cannot infer thresholds or override capacity; it encourages de-escalation when systems are reversible, explainable, and institutionally governed; AI governance mediates the relationship between capability and strategic stability; and it transforms Chicken from a contest of observable resolve to a contest of human-machine credibility and control. The article ends by arguing that for high-stakes strategic interaction, AI governance needs to pay attention not only to accuracy or autonomy, but also to pauseability, interpretability, auditability and credible human restraint.

Keywords: artificial intelligence; game of Chicken; brinkmanship; escalation; commitment; AI governance

1. Introduction

Chicken is a strategy game in which two actors move toward a mutually damaging outcome, each hoping that the other will yield first. Its simple structure is analytically useful because it captures a recurring problem in international politics, military crises, cyber confrontation, economic coercion, and organizational conflict. The actors do not necessarily desire the collision. They want the distributional advantage that comes from the other side yielding first. The danger is then not an accident of the game; it is the medium through which bargaining power is produced. The larger logic is similar to hawk-dove conflict, where players must weigh the value of dominance against the cost of destructive confrontation (Maynard Smith & Price, 1973).

Classical work on conflict, bargaining and signaling explains why rational actors may still approach disaster. During crises, actors can overstate their resolve, generate audience costs, pay sunk costs, raise forces, or commit themselves to make retreat difficult and threats credible (Fearon, 1995, 1997). In Chicken, credibility often comes at the expense of flexibility. The actor who appears least able or willing to turn may force the other actor to turn. But the same logic also produces the central danger of brinkmanship: that commitment can become so persuasive that the road back from the brink narrows or vanishes.

AI changes this logic by changing the instruments through which actors perceive, signal, commit, and de-escalate. Examples of uses for AI-enabled systems include intelligence processing, threat classification, decision support, options recommendation, targeting, cyber defense management, filtering warning indicators, and speeding up decision cycles. Recent research shows that AI already influences military intelligence and targeting (King, 2024), crisis decision making (Horowitz & Lin-Greenberg, 2022), human-autonomy teaming (Mayer, 2023), and trust in high-stakes military contexts (Kreps et al., 2023; Lushenko & Sparrow, 2024). These are important developments for Chicken because the model is not just about preferences, it is about the credibility and interpretation of action under acute time pressure. The problem isn't just that AI makes decisions faster or more rational. The issue is that AI changes the relationship between speed, confidence, delegation and control. But a better early warning system could reduce uncertainty, and it could also tempt actors to move closer to the brink because they think they can quantify risk more precisely. A system that automates response may bolster deterrence. But it may also obscure whether leaders maintain real capacity to pause, revise or reverse action. A recommendation-generating system may improve analysis, but it may also lead to over-reliance, automation bias, or political cover for a hard-line decision (Holbrook et al., 2024; Horowitz & Kahn, 2024). This article develops the concept of algorithmic brinkmanship to account for these changes. Algorithmic brinkmanship is a strategic behavior in which actors use, signal, or rely on AI-enabled perception, prediction, delegation, or automation to shape an opponent’s expectations in a confrontation that has the structure of Chicken. The notion does not imply that machines have independent strategic preferences. Instead, it notes that AI gets integrated into the bargaining apparatus. It influences actors’ beliefs, their speed of action, the credibility of their commitments, and the visibility of their residual ability to restrain themselves.

The article asks: How does AI affect escalation, commitment, and de-escalation in the game of Chicken? The answer is developed in a qualitative conceptual methodology. The analysis draws on peer-reviewed literature. It develops a theoretical model rather than testing new empirical data. This is appropriate, as the aim of the article is to clarify mechanisms and to generate propositions for future empirical research.

2. Research Gap, Objectives, and Research Questions

The gap in the research is not the lack of work on AI and security. That literature has grown rapidly. AI and strategic stability (Altmann & Sauer, 2017; Ayoub & Payne, 2016), AI and deterrence (Johnson, 2020a; Zala, 2024), nuclear instability (Johnson, 2020b), machine delegation (Johnson, 2022), military transformation (Hunter, 2024), human-autonomy teaming (Lyons et al., 2021; Mayer, 2023), military decision support (Horowitz & Lin-Greenberg, 2022), public and elite trust in AI (Kreps et al., 2023; Lushenko & Sparrow, 2024), resort-to-force decision-making (Erskine, 2024; Erskine & Miller, 2024), and the governance of responsible AI (Firlej, 2021; Laux et al., 2024; Papagiannidis et al., 2025; Schraagen, 2023) have been explored by scholars. The scholarship provides valuable insights, but a lot of it considers AI as a generic strategic variable: a source of speed, autonomy, uncertainty or governance risk.

What is left to develop is the exact game-theoretic mechanism by which AI alters brinkmanship in Chicken. Unlike many other models of conflict, Chicken’s bargaining power depends on a paradoxical performance: the actor must look committed enough to make the opponent give way, but not so committed that collision is unavoidable. Existing studies often focus on whether AI stabilizes or destabilizes deterrence. This article asks a sharper question: what impact does AI have on the strategic utility and danger of looking like you can’t turn? This gap matters because AI affects Chicken at the points where the classical model is most fragile. It first changes escalation by reducing the time between seeing and doing. Second, it changes commitment by embedding threats within technical architectures and delegated processes. Third, it alters signaling, since opponents may not know what has been automated, what thresholds are in force, or whether human override is available. Fourth, it changes de-escalation in the sense that exiting the crisis may require technical reversibility, not just diplomatic communication. In the literature, these elements have been considered separately, but not as part of a single account of algorithmic Chicken.

The article aims, therefore, to provide a theoretical account of algorithmic brinkmanship. Specifically, it aims to: (a) clarify how AI affects the escalation logic of Chicken; (b) describe how AI-enabled delegation influences the credibility and control of commitment; (c) explain why AI makes signaling more ambiguous; and (d) specify the conditions under which AI can facilitate rather than hinder de-escalation.

This article is guided by three research questions:

RQ1: How does AI affect escalation dynamics in Chicken-like confrontations?

RQ2: What is the impact of AI on the credibility, interpretation, and reversibility of commitment?

RQ3: Under what conditions can AI help de-escalate rather than escalate brinkmanship?

3. Theoretical framework: bargaining, commitment and control

The theoretical foundation of this article draws on bargaining theory and commitment-problem analysis. Bargaining theory argues that conflict can arise even when there is a mutually preferable settlement because actors have private information, incentives to misrepresent, and problems of credible commitment (Fearon, 1995). Chicken is a very special and vivid form of the problem. The actors want to avoid a collision, but they disagree about who should yield. Each has an incentive to signal resolve while concealing willingness to yield.

The main mechanism is commitment. A commitment is credible when it is costly or difficult to back down. The classical crisis bargaining literature identifies public commitments, audience costs, sunk costs, mobilization, reputation, and irreversible deployments (Fearon, 1997). In Chicken, commitment is created by making one’s own turning less available. But commitment is a two-way street. If it is too weak, the opponent may not give way. If it is too strong, there may not be enough space for either side to de-escalate.

AI falls into this framework in three ways: First, it is an information infrastructure. It affects what actors think they know about the environment, the opponent and the probability of danger. Second, it is an infrastructure of commitment. It can pre-set thresholds, automate responses, prioritize options and reduce the practical time available for humans to reconsider. Third, it is a signaling infrastructure. It can signal readiness, speed, technical sophistication or willingness to delegate action.

The framework thus does not treat AI as a strategically autonomous actor, but as a human-machine decision architecture embedded in bargaining. This is a significant difference. The analytical question is not whether AI has intentions. The question is whether AI alters the credibility and meaning of human and institutional intentions. In a Chicken-like confrontation an AI-enabled posture may tell an opponent: we see faster, we can respond sooner, our response may be predelegated, and your time to influence us is limited. As for the effect of that message, whether it deters, reassures, or provokes depends on how credible, interpretable, and reversible the posture appears.

Table 1. Classical Chicken and Algorithmic Chicken Compared

Dimension	Classical Chicken	Algorithmic Chicken	Strategic implication
Escalation	Visible threats, mobilization, public deadlines, and political resolve.	AI-enabled warning, sensor fusion, decision-support systems, automated cyber or military response.	Escalation becomes faster and may be driven by perceived informational advantage.
Commitment	Credibility through audience costs, sunk costs, mobilization, and reduced freedom to retreat.	Credibility through delegation, pre-set thresholds, system integration, and machine-speed reaction.	Commitment becomes stronger but may become less transparent and less reversible.
Signaling	Signals are interpreted through human intention, reputation, and visible cost.	Signals include technical posture, opacity, automation level, and data-driven readiness.	Opponents may misread defensive automation as offensive preparation.
De-escalation	Negotiation, concession, delay, mediation, and face-saving compromise.	Pause mechanisms, human override, explainable alerts, audit trails, and deconfliction channels.	De-escalation depends on technical reversibility as well as political communication.

Note. The comparison is analytical and synthesizes the theoretical argument; it does not report empirical results.

4. MethodologyThis article uses a qualitative conceptual methodology. Conceptual analysis is appropriate when a research problem needs the clarification of mechanisms, categories, and theoretical relations before systematic empirical testing. This method is particularly well-suited because algorithmic brinkmanship is an emerging phenomenon across the military, cyber, diplomatic, and organizational domains. Direct empirical data on fully fledged AI-enabled Chicken crises are limited, uneven, and often classified. Such a conceptual approach enables the article to formulate theoretically disciplined propositions that can later be tested by case studies, experiments, formal modeling, or simulations.

The analysis proceeds in four steps. First, it extracts from the existing literature of game theory and crisis bargaining the core mechanisms of Chicken: escalation, commitment, signaling, and de-escalation. Second, it sets out how recent AI and security studies describe changes in speed, autonomy, trust, opacity, targeting and governance. Third, it links these AI-related changes to the mechanisms of the Chicken. Fourth, it develops theoretical propositions that specify expected relationships between AI-enabled speed, opacity, delegation, reversibility, and escalation outcomes.

The selection of the literature base was based on three principles. First, the article uses peer-reviewed academic sources with DOIs, which is consistent with the requirements for submission to high-quality journals. Second, it merges classical theoretical work with recent scholarship (2020–2025) so that the argument is grounded in established theory but responsive to current debates. Third, it emphasizes sources that address issues related to strategic decision-making, human-machine trust, autonomous systems, AI governance, and crisis behavior, rather than general discussions of digital technology.

The methodology is analytical, not empirical. It does not claim to measure the frequency of algorithmic brinkmanship, or to test the propositions statistically. Its validity is based on theoretical consistency, clear conceptual representation, and consistency with current peer-reviewed findings. This design is common in early stage theory building, where the objective is to render a phenomenon researchable by specifying its mechanisms, scope conditions, and observable implications.

The paper is focused on strategic interactions with Chicken-type payoffs. It does not claim all AI-enabled conflict is like Chicken. Some interactions resemble Prisoner’s Dilemma, Stag Hunt, bargaining over indivisible goods, or repeated deterrence games. The argument is most directly relevant to situations in which two actors have an interest in avoiding mutual harm but are competing over who must yield, delay, or concede first.

Figure 1. Mechanism model of algorithmic brinkmanship in Chicken-like interaction.

5. Analysis: Escalation Tempo and Algorithmic Confidence

Escalation in Chicken is not just movement toward conflict, it is movement toward danger for bargaining effect. The actor escalates to make the opponent think that the costs or futility of continuing resistance are high. AI changes escalation by increasing the speed, volume and apparent precision of information. Military and security organizations are increasingly using AI to analyze intelligence, assist targeting, prioritize warnings and structure choices (King, 2024). These systems can improve awareness, but they also change the pace of crisis interaction.

Mechanism one is time compression. Rapidly detecting signals and suggesting responses may put pressure on leaders to act before the opponent gains the upper hand. But that does not mean the death of human judgement. Instead, judgment is increasingly made in compressed time and system-generated urgency. In a Chicken-like crisis, time compression is dangerous because the actors are already trying to convince each other that they will not be the first to swerve. The quicker the interaction, the less scope for clarification, mediation or face-saving adjustment.

The second mechanism is confidence inflation. AI can give the impression that risk is measurable more precisely than it is. The person making a decision when given a probability estimate, a pattern classification or an optimized recommendation may feel that escalation can be better managed. Better information can mean better action in everyday management contexts. But in brinkmanship, confidence can breed risk-taking. Leaders who think AI will find the last safe chance to turn may drive to the point of collision.

The third mechanism is adversarial interpretation. Horowitz and Lin-Greenberg (2022) demonstrate how the use of AI can influence how national security experts understand crisis events, including rival accidents involving AI-enabled systems . This matters for Chicken, as the same behavior can be interpreted as resolve, error, loss of control, or preparation to attack. AI does not eliminate ambiguity. It may shift ambiguity from human intention to system behavior. One might ask of an opponent: Was this move deliberate? Was it triggered by a threshold? Can it be reversed? Who can stop it?

So AI can reduce certain forms of uncertainty and increase others. It might improve environmental knowledge but reduce social and strategic interpretation. Ignorance alone is not the core risk of escalation. The core risk is misplaced confidence under ambiguous interdependence.

6. Analysis: Commitment, Delegation, and the New Steering Wheel

The classical metaphor of Chicken often involves throwing away the steering wheel. The gesture communicates that the actor can no longer turn, forcing the opponent to choose between concession and collision. AI creates new forms of steering-wheel removal. Predelegated response systems, automated cyber defenses, predictive targeting pipelines, and machine-speed warning architectures can all reduce the apparent role of discretionary human choice.

Delegation can strengthen commitment because it makes response appear less dependent on political hesitation. Johnson (2022) argues that delegating strategic decision-making to machines raises serious questions about stability, escalation, and control. In Chicken, this matters because a threat becomes more credible when the opponent believes the actor cannot easily back away. An automated threshold may function as a technical red line. If crossed, it may generate a response with limited delay.

Yet algorithmic commitment differs from classical commitment in an important way. It is often difficult for outsiders to verify. Public commitment can be seen and heard; algorithmic commitment may be hidden in architecture, code, data pipelines, command rules, or organizational practice. An actor may exaggerate automation to appear resolute, conceal automation to preserve advantage, or misunderstand its own system's practical rigidity. This creates a credibility-interpretability gap: commitment may be strong but not legible, or legible but not truly strong.

This gap matters because deterrence requires communication. A red line that cannot be understood may not deter; it may only surprise. A threshold that changes dynamically may be difficult for the opponent to avoid. A response system that lacks visible human override may convince the opponent that communication is useless. Thus, AI-enabled commitment can simultaneously increase credibility and reduce crisis manageability.

The commitment problem is also organizational. AI outputs can become political resources. Leaders may use system recommendations to justify hard-line positions or to resist compromise. Once a decision is framed as technically validated, retreat may appear irrational, weak, or irresponsible. In this sense, AI can create internal audience costs. A leader may become tied not only to a public threat but also to the authority of a system that has been presented as objective.

7. Analysis: Signaling, Opacity and Trust

Chicken signaling is predicated on the opponent’s ability to infer resolve from action. AI makes this inference complicated. A technical deployment could be a sign of defensive vigilance, offensive preparation, bureaucratic modernization, political theater, or a true willingness to automate escalation. Opponents may not know what interpretation is right. Recent work on trust and automation can help explain this ambiguity, not just external but also internal. Human operators may overtrust or undertrust AI depending on system performance, task design, stakes, and institutional culture (Dietvorst et al., 2015; Logg et al., 2019; Parasuraman & Riley, 1997) . Experimental work indicates that trust in AI is context-dependent and can differ based on purpose, oversight, precision, and perceived risk (Kreps et al., 2023; Lushenko & Sparrow, 2024; Mayer, 2023). Holbrook et al. (2024) illustrate the danger of overtrust in life-and-death recommendations. These results suggest that algorithmic Chicken is not just a game of two rational calculators. It’s a competition between organizations that may not be uniformly dependent on machines, and whose machine dependence may be contested and hard for adversaries to read.

The opponent's belief about trust becomes strategic. If an actor is thought to overtrust AI, its threats may seem more dangerous because it may act rigidly or prematurely. If an actor is perceived as not trusting AI, its technical posture could be less credible. If an actor publicly claims human control but privately relies heavily on automated pipelines, the signal could be unstable. The opponent must read not only intentions but the human-machine relationship within the rival’s decision architecture.

Here is where AI governance becomes immediately relevant to game theory. Governance principles like transparency, accountability, human oversight, auditability and risk management are not only ethical or legal concerns. They are strategic variables in Chicken-like interaction. Laux et al. (2024) caution against equating trustworthiness with acceptability of risk. Papagiannidis et al. (2025) define responsible AI governance as structural, relational, and procedural. These governance dimensions shape whether opponents can understand thresholds, believe in restraint, and find de-escalatory paths within algorithmic brinkmanship.

Figure 2. Strategic zones created by speed, opacity, and reversibility in algorithmic Chicken.

8. Theoretical Propositions

The analysis can be summarized in six theoretical propositions. These propositions are not statistical results. They are theory-building claims that specify observable relationships for future research.

Proposition 1: The risk of escalation in Chicken-like crises is increased by AI-enabled speed when the capacity to compress deliberation outstrips the capacity to improve shared interpretation. The main risk is not speed itself but speed plus ambiguity. When machine outputs accelerate action without generating mutual understanding, actors may drift toward collision before political communication can remedy misperception .

Proposition 2: When actors believe predictive systems allow them to manage danger more precisely than the strategic environment permits, AI-enabled confidence increases brinkmanship. This proposition suggests a confidence / risk route. Better data can reduce uncertainty, but it can also lead actors to accept more danger by overestimating their control.

Proposition 3: When opponents believe that response thresholds are automatic or difficult to reverse, AI-enabled delegation of response decisions enhances the credibility of commitment. This generalizes classical commitment theory. Automation can be like a technical audience cost or a digital steering wheel that is discarded.

Proposition 4: AI-enabled commitment is destabilizing when the thresholds, override rules, or accountability structures are opaque to the opponent. Commitment has to be credible, but it has to be comprehensible. Opaque commitment may not deter, but it raises the risk of accidental or inadvertent escalation.

Proposition 5: AI aids de-escalation when systems are built with reversibility, explanation and credible human override. Algorithmic Chicken de-escalation takes more than a diplomatic note. It requires the practical ability to stop technical processes, audit machine recommendations and communicate that restraint can still be exercised.

Proposition 6: AI governance mediates the relationship between AI capability and strategic stability. The same AI capacity could be stabilizing or destabilizing depending on the design of institutions. Governance is thus not exogenous to the game, but changes the payoff-relevant beliefs that actors have about credibility, control and restraint.

Table 2. Propositions and Observable Implications

Proposition	Core mechanism	Observable implication for future research
P1: Speed-risk proposition	AI compresses the perception-action cycle.	Crises with faster AI-enabled warning and response should show shorter windows for diplomacy and higher reliance on preplanned moves.
P2: Confidence-risk proposition	AI creates perceived precision and control.	Actors with high confidence in prediction tools should be more willing to escalate close to thresholds.
P3: Delegated-commitment proposition	Automation makes threats appear less discretionary.	Public or inferred automation should increase perceived resolve, especially when thresholds appear pre-set.
P4: Opacity-instability proposition	Opaque thresholds weaken shared interpretation.	Ambiguous AI posture should increase misperception and reduce the deterrent value of signals.
P5: Reversibility proposition	Pauseability and override preserve exits.	Systems with clear human override and explainability should lower escalation persistence after warning errors.
P6: Governance-moderation proposition	Rules shape trust and strategic interpretation.	Governance practices should moderate whether AI-enabled capabilities are read as stabilizing or threatening.

Note. The propositions are theoretical claims derived from the conceptual analysis and are intended for future empirical testing.

9. Discussion: Implications for Game Theory, Strategic Studies, AI Governance

9.1 Contributions to Game Theory

The article contributes to game theory by demonstrating that the standard Chicken model should be extended to include decision architecture. Classical Chicken is about preferences and strategic choices: go on or swerve. This article introduces a prior and concurrent layer: how the actors perceive the road, how quick they are to react, how their commitments are technically embedded and whether the opponent can understand their residual capacity to turn. The contribution is not to replace Chicken, but to refine its assumptions for AI-enabled interaction.

Specifically, the article points to credibility, interpretability and reversibility as related variables. Credibility often dominates classic accounts: An actor who can make retreat costly may win. Algorithmic Chicken shows that credibility without interpretability can go wrong. If the adversary can’t discern what’s been automated, or where thresholds are, commitment may not be a sign of resolve. Credibility without reversibility may produce collision rather than bargaining success, similarly. Game-theoretic models of Chicken should therefore take into account the legibility and pauseability of commitment devices, in addition to their strength.

9.2 Contribution to Strategic Studies

The article explains why the impact of AI on strategic stability cannot be evaluated only in terms of capabilities. AI-enabled systems could improve intelligence, targeting, warning and coordination but their strategic impact will depend on the context of a crisis. In Chicken-like encounters, capabilities that seem operationally efficient can be strategically dangerous, if they compress time, obscure intention, or reduce space for controlled retreat. This is why the same AI system can be stabilizing in routine monitoring but destabilizing in a crisis.

The article also redefines deterrence and brinkmanship. Deterrence is more than just capability and resolve. It also demands that adversaries know what actions will elicit response and whether communication can still change outcomes. Algorithmic brinkmanship thus expands strategic studies from a narrow focus on autonomous weapons to a broader focus on AI-enabled decision architectures. And even if they are not weapons, targeting systems, intelligence processing, cyber defense, logistics and warning tools can all have an effect on escalation.

9.3 Contribution to AI Governance

The article demonstrates that governance principles have strategic effects for AI governance. Ethical or legal safeguards are often raised in the form of human oversight, transparency, explainability, auditability and accountability. They also serve as de-escalation mechanisms in the game of Chicken. A system that can be paused, explained and overridden, signals something different from a system that appears automatic, opaque and irreversible.

This contribution is important because many AI governance frameworks are concerned with internal risk management: Is the system accurate, fair, accountable, and compliant? Algorithmic brinkmanship introduces an external relational requirement: can adversaries, partners, and crisis interlocutors establish reliable beliefs about the thresholds of the system and human control? So governance is not just about keeping AI safe within an organization. It’s also about making AI behavior strategically interpretable to others when the stakes are high.

10. Limitations and Future Research

The article has a few limitations. First, it is conceptual, not empirical. It develops mechanisms and propositions but doesn’t test them against a dataset or case archive. This is appropriate for theory building, but the propositions need systematic empirical evaluation.

Second, the article treats AI at an abstract level that covers decision support, warning, targeting, cyber response, and autonomous functions. Future research should disaggregate these technologies, as different systems might have different effects on escalation.

Third, the article is focused on Chicken-like interactions, and does not claim that all AI-enabled crises have this structure. Not all conflicts are Chicken. Some involve repeated bargaining, alliance reassurance, arms racing, or cooperation problems. Future work should compare the effects of AI across games.

Fourth, the article does not give a formal mathematical model. Formal modeling could help to clarify equilibrium conditions under different assumptions on speed, opacity and reversibility.

Four directions for future research are proposed. Case studies could explore crises involving automated warning, cyber defense, drone escalation, or AI-enabled targeting to determine whether the mechanisms suggested here are present in practice. Experimental research could test whether decision-makers are more prone to escalate when they are supported by high-confidence algorithmic advice. Wargaming could look at how opponents interpret different levels of AI delegation versus human override. Finally, governance research could examine what transparency and pauseability measures are credible enough to support de-escalation without revealing sensitive capabilities.

11. Conclusion This article has argued that AI changes the game of Chicken by changing the mechanisms of escalation, commitment, signaling and de-escalation. The primary contribution is the idea of algorithmic brinkmanship: the use, presentation, or reliance upon AI-enabled perception, prediction, delegation, and automation to shape an opponent’s expectations under mutual peril.

The article demonstrates that AI does not simply make strategic actors more rational or more reckless. But its effect depends on the way in which speed, opacity, delegation and reversibility are combined. AI can improve warning and support restraint but can also shorten deliberation, inflate confidence, harden commitment, and make signals more difficult to interpret. The game is therefore not a contest of visible will, but a contest of human-machine credibility and control.

For theory, the article extends Chicken by adding decision architecture to commitment analysis. For strategic studies, it provides an explanation of why AI-enabled capability can be operationally useful and strategically destabilizing. For AI governance, it shows that oversight, explanation, auditability and human override are not only ethical safeguards; they are de-escalation mechanisms. The practical lesson is simple: in high-stakes strategic interaction, AI systems need to be designed not only to act, but to pause, explain, and leave room to turn away.

References

Altmann, J., & Sauer, F. (2017). Autonomous weapon systems and strategic stability. Survival, 59(5), 117-142. https://doi.org/10.1080/00396338.2017.1375263

Ayoub, K., & Payne, K. (2016). Strategy in the age of artificial intelligence. Journal of Strategic Studies, 39(5-6), 793-819. https://doi.org/10.1080/01402390.2015.1088838

Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114-126. https://doi.org/10.1037/xge0000033

Erskine, T. (2024). Before algorithmic Armageddon: Anticipating immediate risks to restrain when AI infiltrates decisions to wage war. Australian Journal of International Affairs, 78(2), 175-190. https://doi.org/10.1080/10357718.2024.2345636

Erskine, T., & Miller, S. E. (2024). AI and the decision to go to war: Future risks and opportunities. Australian Journal of International Affairs, 78(2), 135-147. https://doi.org/10.1080/10357718.2024.2349598

Fearon, J. D. (1995). Rationalist explanations for war. International Organization, 49(3), 379-414. https://doi.org/10.1017/S0020818300033324

Fearon, J. D. (1997). Signaling foreign policy interests: Tying hands versus sinking costs. Journal of Conflict Resolution, 41(1), 68-90. https://doi.org/10.1177/0022002797041001004

Firlej, M. (2021). Regulating human control over autonomous systems. Regulation & Governance, 15(4), 1071-1088. https://doi.org/10.1111/rego.12344

Holbrook, C., Holman, D., Clingo, J., & Wagner, A. R. (2024). Overtrust in AI recommendations about whether or not to kill: Evidence from two human-robot interaction studies. Scientific Reports, 14, Article 19751. https://doi.org/10.1038/s41598-024-69771-z

Horowitz, M. C., & Kahn, L. (2024). Bending the automation bias curve: A study of human and AI-based decision making in national security contexts. International Studies Quarterly, 68(2), sqae020. https://doi.org/10.1093/isq/sqae020

Horowitz, M. C., & Lin-Greenberg, E. (2022). Algorithms and influence: Artificial intelligence and crisis decision-making. International Studies Quarterly, 66(4), sqac069. https://doi.org/10.1093/isq/sqac069

Hunter, C. (2024). We'll never have a model of an AI major-general: Artificial intelligence, command decisions and kitsch visions of war. Journal of Strategic Studies, 47(4), 685-706. https://doi.org/10.1080/01402390.2023.2241648

Johnson, J. (2020a). Deterrence in the age of artificial intelligence and autonomy: A paradigm shift in nuclear deterrence theory and practice? Defense & Security Analysis, 36(4), 422-448. https://doi.org/10.1080/14751798.2020.1857911

Johnson, J. (2020b). Artificial intelligence in nuclear warfare: A perfect storm of instability? The Washington Quarterly, 43(2), 197-211. https://doi.org/10.1080/0163660X.2020.1770968

Johnson, J. (2020c). Artificial intelligence, drone swarming and escalation risks in future warfare. The RUSI Journal, 165(2), 26-36. https://doi.org/10.1080/03071847.2020.1752026

Johnson, J. (2022). Delegating strategic decision-making to machines: Dr. Strangelove redux? Journal of Strategic Studies, 45(3), 439-477. https://doi.org/10.1080/01402390.2020.1759038

King, A. (2024). Digital targeting: Artificial intelligence, data, and military intelligence. Journal of Global Security Studies, 9(2), ogae009. https://doi.org/10.1093/jogss/ogae009

Kreps, S., George, J., Lushenko, P., & Rao, A. (2023). Exploring the artificial intelligence trust paradox: Evidence from a survey experiment in the United States. PLOS ONE, 18(7), e0288109. https://doi.org/10.1371/journal.pone.0288109

Laux, J., Wachter, S., & Mittelstadt, B. (2024). Trustworthy artificial intelligence and the European Union AI Act: On the conflation of trustworthiness and acceptability of risk. Regulation & Governance, 18(1), 3-32. https://doi.org/10.1111/rego.12512

Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90-103. https://doi.org/10.1016/j.obhdp.2018.12.005

Lushenko, P., & Sparrow, R. (2024). Artificial intelligence and U.S. military cadets' attitudes about future war. Armed Forces & Society. Advance online publication. https://doi.org/10.1177/0095327X241284264

Lyons, J. B., Sycara, K., Lewis, M., & Capiola, A. (2021). Human-autonomy teaming: Definitions, debates, and directions. Frontiers in Psychology, 12, Article 589585. https://doi.org/10.3389/fpsyg.2021.589585

Mayer, M. (2023). Trusting machine intelligence: Artificial intelligence and human-autonomy teaming in military operations. Defense & Security Analysis, 39(4), 521-538. https://doi.org/10.1080/14751798.2023.2264070

Maynard Smith, J., & Price, G. R. (1973). The logic of animal conflict. Nature, 246, 15-18. https://doi.org/10.1038/246015a0

Papagiannidis, E., Mikalef, P., & Conboy, K. (2025). Responsible artificial intelligence governance: A review and research framework. Journal of Strategic Information Systems, 34(2), Article 101885. https://doi.org/10.1016/j.jsis.2024.101885

Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), 230-253. https://doi.org/10.1518/001872097778543886

Schraagen, J. M. (2023). Responsible use of AI in military systems. Ergonomics, 66(12), 1905-1913. https://doi.org/10.1080/00140139.2023.2278394

Zala, B. (2024). Should AI stay or should AI go? First strike incentives and strategic stability. Australian Journal of International Affairs, 78(2), 200-210. https://doi.org/10.1080/10357718.2024.2328805

#Artificial_Intelligence #Executive_Education #Dynamic_Capabilities #Leadership_Development #Strategic_Learning #Organizational_Adaptation #Algorithmic_Governance