Algorithmic Statism: A New Security Model in the US Artificial Intelligence Strategy and Its Global Implications

ARTIFICIAL INTELLIGENCE IN TECHNOLOGY GOVERNANCE AND THE REDEFINITION OF STATE POWER

For centuries, international relations theory has defined state power in terms of territory, population, military strength and raw material reserves. The fundamental axioms of the Westphalian system established an analytical framework that linked sovereignty to physical geography and explained power competition in terms of material capacities. With the rise of the knowledge economy in the final quarter of the twentieth century, this framework has largely been eroded; Joseph Nye’s concept of ‘soft power’ and Manuel Castells’ theory of the ‘network society’ have argued that power relations are shaped not so much by material factors as by connectivity, information flows and the capacity to narrate. However, the transformation taking place today is leaving even these second-generation approaches behind: artificial intelligence is not merely altering economic productivity or public communication; it is reshaping state power itself, weapons systems, intelligence mechanisms, critical infrastructure and even institutional decision-making capacity.

One of the most striking pieces of evidence for this transformation is a series of two developments that took place in the US in June 2026. The White House first imposed export restrictions on Anthropic’s Mythos 5 and Fable 5 models on national security grounds; whilst partially relaxing this decision, it announced that access to these models would be restricted to over a hundred American organisations and companies deemed trustworthy. Subsequently, it emerged that the Trump administration had demanded OpenAI halt the general release of its GPT-5.6 model and, initially, make it available only to a limited number of government-approved partners. This decision sets a historic precedent in that it marks the first time the US government has directly intervened in the distribution of an artificial intelligence model that has not yet been released to the market.

These developments cannot be viewed merely as administrative decisions concerning two technology companies. On the contrary, they represent the extension of the theory of ‘armed interdependence’ to the algorithmic realm; the state’s search for a new equilibrium in its relationship with technology companies and market forces; and the confirmation that artificial intelligence is now being managed as a dual-use geostrategic asset.

ALGORITHMIC STATISM: CONCEPTUAL FRAMEWORK AND HISTORICAL BACKGROUND

The concept of ‘algorithmic statism’ is employed in this article with a specific theoretical connotation: it refers to the phenomenon whereby states establish systematic control and steering mechanisms over the development, distribution and access to artificial intelligence models, on the grounds of national security and strategic superiority. This concept points to a different dimension from the concepts of “data sovereignty” and “digital authoritarianism” previously discussed in the literature: what is at issue here is not merely controlling data flows or transforming digital infrastructure into a tool of political coercion. The crux of the matter is classifying algorithmic capacity itself—much like nuclear weapons or biological agents—as a strategic force multiplier that must be controlled by the state.

To understand the historical roots of this concept, it is illuminating to look at the management of dual-use technology during the Cold War era. The EAR (Export Administration Regulations) regime and the ITAR (International Traffic in Arms Regulations) framework, which govern US export controls, have for decades subjected certain technology categories to licensing requirements on the grounds that they can be used for both civilian and military purposes. Semiconductors, precision optical systems and cryptographic technologies are among the historical examples of this dual-use control regime. The US’s restriction in 2022 on the export of next-generation semiconductors and GPUs to China has been the most recent and comprehensive manifestation of this historical process.

The inclusion of artificial intelligence models on this list, however, represents a qualitatively new phase. A GPU is a physical product; its manufacture, replication and control are, to a certain extent, framed by tangible physical constraints. By contrast, an advanced large language model is an abstract entity existing in the form of weight parameters and software code; it can be digitally replicated, transferred and, to some extent, reproduced through reverse engineering. Consequently, algorithmic statism presents a far more complex and paradoxical governance challenge than control over hardware. Indeed, the Trump administration initially repealed rules—inherited from the Biden era—that mandated security reviews of leading models; however, in June 2026, it effectively shifted towards a far more direct intervention model with the GPT-5.6 and Mythos restrictions, and this inconsistency has led to a climate of serious uncertainty within the US sector.

THE GPT-5.6 AND MYTHOS CRISIS: ANATOMY OF THE CONDITIONAL LICENSING MODEL

In June 2026, the Office of the National Cyber Director and the Office of Science and Technology Policy, both attached to the White House, requested that OpenAI roll out GPT-5.6 in a controlled and phased manner. According to information reported by Turkey-based news sources such as Karar and Harici, citing the US press, OpenAI CEO Sam Altman stated internally that the model would initially be shared only via the Amazon Bedrock platform with approximately twenty government-approved organisations. Altman clearly emphasised that this restrictive approach is not OpenAI’s preferred long-term method, stating: “We do not believe that this kind of government access process should become the default norm in the long term.”

The Anthropic case, which occurred immediately prior to this decision, highlights that the trend towards algorithmic statism is taking on a systematic dimension. On 13 June 2026, the Trump administration imposed export restrictions on Anthropic’s Mythos 5 and Fable 5 models on national security grounds; whilst partially relaxing these restrictions, it announced that access to these models would be limited to over a hundred US organisations and companies deemed trustworthy. It is reported that Commerce Secretary Howard Lutnick opposed the release of even a limited version of GPT-5.6, calling CEO Altman directly to demand approval from other federal agencies. This information indicates that the process in question extends not only to security agencies but also to the highest echelons of economic management.

The technical rationale behind these decisions lies in the advanced capabilities of both models in the field of cyber security. As analysts have pointed out, the GPT-5.6 Sol model is capable of delivering significantly higher performance than previous-generation systems in detecting and patching security vulnerabilities in digital infrastructure. When used in the opposite direction, this same capability could also facilitate the development of autonomous cyber-attack tools. This dual-use potential places the model in a category functionally similar to that of fissile material or chemical weapon precursors; consequently, it transforms state regulation from a purely ethical concern into a concrete security necessity.

However, the internal contradictions of the decision should not be overlooked. To put it in geo-economic terms: the transformation of trade policy into a security tool could threaten both the global competitiveness of the companies concerned and US technological leadership. The fact that Silicon Valley—a traditional supporter of free-market technology policy—has begun to re-examine its support for the Trump administration in the context of these restrictions; the emergence of a landscape of uncertainty—described by Politico as an ‘unfettered and bewildering regulatory environment’—and the fact that some industry executives are nostalgically comparing this approach to the more balanced stance of the Biden era, are the most evident manifestations of this internal contradiction.

THE ALGORITHMIC DIMENSION OF THE US-CHINA AI RIVALRY

The deepest global repercussions of the GPT-5.6 and Mythos crisis are undoubtedly unfolding in this new phase of the US-China technology rivalry. To make sense of this competition, it is illuminating to refer to a framework of complex interdependence: the global technology ecosystem has, over the past decade, become increasingly interwoven with an asymmetrical relationship of interdependence; however, this relationship of interdependence is now generating mutual vulnerabilities rather than mutual gains, and is taking on a ‘weaponised’ character.

Viewed through this lens, the strategy pursued by the US is seen to encompass a three-tiered logic. The first tier consists of hardware restrictions: the export controls on next-generation semiconductors and GPUs, implemented from 2022 onwards, aim to limit China’s capacity to produce the processing power required for artificial intelligence. The second layer consists of software restrictions: access restrictions imposed on the GPT-5.6 and Mythos models are designed to prevent Chinese state actors and institutions from utilising the most advanced algorithmic systems. The third layer consists of the increasingly pronounced competition over standard-setting: the question of who sets AI security standards, ethical frameworks and international governance norms is taking on a critical dimension in terms of the reproduction of the hegemonic order.

China’s response to this strategy, however, constitutes a structural transformation. Every new US restriction serves to strengthen Beijing’s motivation to accelerate its domestic AI ecosystem. The rapid progress observed in large language models, the massive investments in semiconductor production infrastructure, and efforts to build algorithmic capacity through open-source models are concrete indicators of this reactive dynamism. Given that one of Washington’s concerns is that China might replicate American AI capabilities through reverse engineering, cyber espionage or open-source models, algorithmic statism serves as both an offensive and a defensive strategy.

This competition also carries the risk of deepening technological polarisation within the international system. On the one hand, there are the US models—state-directed yet formally sourced from the private sector and subject to security vetting; on the other, China’s systems, which are directly state-funded and rely on ideologically controlled data pools. Other actors caught between these two poles must confront strategic choices regarding which ecosystem to align with.

GLOBAL REGULATORY DIVERGENCE AND THE CLASH OF GOVERNANCE PARADIGMS

The trend towards algorithmic statism does not stem solely from the US’s individual policy preferences; it is part of a broader global regulatory transformation. To make sense of this transformation, it is necessary to evaluate three fundamental governance paradigms in a comparative manner.

The first paradigm is the ‘conditional access’ model adopted by the US. Under this model, artificial intelligence systems are subject to national security scrutiny once they exceed a certain capacity threshold. Access is not completely blocked; however, conditional authorisation is granted to a limited number of actors whose reliability has been assessed. The advantage of this approach is that it does not completely stifle innovation; the disadvantage, however, is the uncertainty of its legal framework and, as Politico has also highlighted, the unpredictable nature of its implementation. An executive order signed by the Trump administration in June 2026 stipulates that powerful models must be submitted for voluntary federal review for up to thirty days prior to their market launch; however, a clear framework governing the operation of this system has not yet been established.

The second paradigm is the ‘risk-based regulation’ model adopted by the EU. The European Union’s Artificial Intelligence Act (AI Act) classifies models into categories of unacceptable risk, high risk, limited risk and minimal risk according to their areas of use; it imposes strict transparency, auditability and compliance obligations for high-risk applications. The key difference in this approach is that oversight lies with independent regulatory bodies rather than security agencies, and the framework is grounded in a transparent legal basis. However, the EU model also has a significant weakness: the burden of bureaucracy and high compliance costs carry the risk of eroding the competitiveness of European companies in the global AI race.

The third paradigm is China’s ‘party-led oversight’ model. Under this model, the development and deployment of artificial intelligence are conducted within a framework of state planning aligned with the Communist Party’s ideological and security interests. Content filtering obligations, data localisation requirements and comprehensive restrictions on algorithmic recommendation systems are among the most prominent features of this model. Whilst this paradigm is strong in terms of rapid scaling and the funding of national champions, its weakness lies in the fact that ideological filtering limits the quality of research and global competitiveness in the long term.

The tension between these three paradigms will constitute the most critical area of normative conflict in international AI governance over the next decade. AI governance initiatives within the G7, G20 and the UN fall short of establishing universally binding standards; each of the three power centres is seeking to build techno-economic spheres of influence that will effectively make their own norms dominant.

THE INNOVATION DILEMMA: THE TENSION BETWEEN STATE REGULATION AND TECHNOLOGICAL LEADERSHIP

The deepest internal contradiction of algorithmic statism lies in the potential conflict between the goal of safeguarding state security and that of maintaining technological leadership. This tension can be viewed as a technological variant of the ‘security dilemma’ in the literature on international relations: every restrictive measure taken for defence purposes may increase rival actors’ motivation to develop their own independent capabilities, thereby deepening rather than alleviating security concerns.

This dynamic has been observed in a particularly striking manner in the case of GPT-5.6. The restrictions imposed on OpenAI by the Trump administration have called into question both the sector’s internal dynamics and the free-market support that US technology companies traditionally expect from a Republican administration. According to a report in the Harici newspaper, a senior industry executive described this measure as ‘effectively resembling a European-style licensing regime’; other executives, meanwhile, expressed a nostalgic view of the Biden administration’s more predictable approach. This is not merely a matter of rhetoric; the climate of uncertainty has the potential to adversely affect long-term research and development decisions.

However, it must be acknowledged that a complete rejection of algorithmic statism is not a rational option either. In a world where advanced artificial intelligence models have genuinely acquired a dual-use character, simultaneously enhancing both cyber defence and cyber attack capabilities, for liberal democratic states to leave these systems entirely to market dynamics could amount to taking an unacceptable security risk. The experience of nuclear governance during the Cold War has demonstrated that, whilst this tension can never be fully resolved, it is at least possible to make it manageable: export control regimes, disarmament negotiations and bilateral confidence-building mechanisms are among the elements of this governance repertoire.

The question of whether a similar governance framework can be established in the field of artificial intelligence constitutes one of today’s most critical global governance issues. The artificial intelligence security summit hosted by the United Kingdom in 2023 and the subsequent Seoul Declaration have sent encouraging signals regarding the will to establish a multilateral framework in this regard; however, the deep strategic mistrust between the US and China continues to stand as a major obstacle to the establishment of a binding international regime.

TURKEY’S POSITION: A MIDDLE-POWER STRATEGY IN THE AGE OF ALGORITHMIC STATISM

The deepening of the trend towards algorithmic statism on a global scale creates both concrete threats and unique opportunities for middle-power states such as Turkey. To assess this situation, it would be appropriate to use the theory of middle powers as a framework: according to this theory, middle powers can enhance their effectiveness by engaging in multilateral diplomacy, utilising niche expertise and acting as bridge-builders within the structural gaps created by great power competition.

Viewed through this lens, the threats facing Turkey are quite tangible. Firstly, should access restrictions intensify, Turkey could potentially be excluded from the most advanced artificial intelligence systems; this situation could entail the risk of being deprived of significant capabilities in terms of defence, critical infrastructure and competitive economic development. Secondly, algorithmic statism also signifies a deepening of the sovereignty issue currently debated in the context of data centre investments: the question of under which legal framework and by which actors AI models are operated is no longer merely a commercial matter; it has taken on a directly strategic dimension. The limited capacity of Turkey to develop its own large language model is, in this context, turning into a structural vulnerability.

However, the opportunities should not be overlooked. Turkey’s capacity to maintain bridging relationships with both the US and the EU simultaneously could acquire unique value in the age of algorithmic statism. Turkey, which could serve as a normative bridge between the US’s conditional access model and the EU’s risk-based regulatory model, can position itself in strategic niches such as cyber security cooperation, a trusted access partnership and the role of a regional AI hub.

Three policy priorities stand out for realising this potential. The first priority is to accelerate the process of aligning with the EU AI Act and finalise the GDPR adequacy decision: this step will serve as a critical signal to elevate Turkey to the status of a ‘trusted partner’ in the eyes of both the EU and the US. The second priority is the systematic implementation of a state-supported programme to develop a domestic large language model: the most fundamental guarantee of strategic independence lies in possessing algorithmic capacity; without this capacity, any rhetoric on data sovereignty will ring hollow. The third priority is to upgrade cyber security capabilities to international standards and to take an active role in multilateral platforms in the field of artificial intelligence security: This area could be regarded as a niche field in which Turkey can both reinforce its credibility and enhance its normative influence.

CONCLUSION: FROM THE NUCLEAR AGE TO THE ALGORITHMIC AGE

Nuclear technology, which profoundly shook the international system in the mid-twentieth century, was ultimately transformed into a manageable threat by subjecting the knowledge and materials required to produce the atomic bomb to inter-state control mechanisms. This process was neither swift nor smooth: the journey from Hiroshima to the NPT (Treaty on the Non-Proliferation of Nuclear Weapons) took approximately a quarter of a century; it required a shared perception of threat, forged by great power rivalry and the fear of imminent destruction.

Artificial intelligence, however, now stands on the brink of a similar governance crisis. The US’s decisions regarding GPT-5.6 and Mythos in June 2026 will go down in history as the official declaration that this crisis has, in fact, begun. The fact that states have begun to classify algorithmic capacity—much like nuclear knowledge—as a strategic power multiplier that cannot be left solely to the market; and the permanent shift from the ‘develop first, regulate later’ paradigm to the ‘assess safety first, then grant access’ approach, represents one of the most enduring turning points in history.

However, just as establishing a presence at the junctions of algorithmic networks generates power, severing these networks can lead to an erosion of power. The US may achieve short-term security gains by establishing control over OpenAI and Anthropic; yet, due to the unpredictability of this control model and the fact that it may bolster China’s motivation to develop domestic capabilities, it also carries the risk of eroding its own technological superiority in the long term. This paradox constitutes the fundamental tension of the age of algorithmic statism: control is necessary to safeguard security; yet control may threaten the superiority in innovation that is the very foundation of that security.

For Turkey, the lesson is clear: in the algorithmic age, sovereignty will be secured not only through territorial integrity, but also through domestic algorithmic capacity, the status of a trusted partner, and an active role in international normative processes. Developing an integrated strategy encompassing these three dimensions will continue to be the key factor determining Turkey’s position in this increasingly intense geostrategic competition.