On June 27, 2026, the US government partially restored access to Anthropic’s Mythos 5 model for a defined population of “trusted” companies and agencies, fifteen days after suspending it entirely. The episode, and the parallel staggered rollout the same week of OpenAI’s GPT-5.6, mark something more consequential than a regulatory dispute resolved. They mark the operational debut of frontier AI access as a tool of American international influence — a lever whose grant, suspension, and tiered restoration the United States is learning to wield with the same deliberateness it has historically applied to arms sales, technology transfer, and intelligence sharing.
By Vladimir Tsakanyan, PhD · Center for Cyber Diplomacy and International Security · cybercenter.space
The history of American technological statecraft has a recognisable pattern. A capability emerges in which the United States holds a decisive advantage. The government develops, often through trial and improvisation rather than through a pre-existing doctrine, a framework for deciding who may access that capability, under what conditions, and with what leverage attached to the granting or withholding of access. Nuclear technology sharing under the Atoms for Peace programme, satellite and missile technology controls under the Missile Technology Control Regime, semiconductor export restrictions under the post-2019 China technology controls — each followed this trajectory: capability advantage, improvised control mechanism, eventual formalisation into a doctrine of access-as-leverage.
The events of June 2026 mark the point at which frontier AI joined this lineage. The two-week Mythos 5 suspension and its replacement with a tiered “trusted” access framework was not, in its final form, primarily a cybersecurity remediation exercise. It was the United States government demonstrating, to itself and to every observer with the capacity to notice, that it now possesses the technical and institutional capacity to grant, restrict, and selectively reinstate access to the most capable AI systems in the world — and that this capacity functions as a form of leverage exercisable over both domestic companies and the foreign governments and organisations that depend on American frontier AI capability.
The Mechanics of the New Leverage
The trusted-tier framework that emerged from the Mythos 5 negotiation establishes, for the first time in operational form, the mechanism through which this leverage functions. Access to a frontier AI model with demonstrated expert-level cybersecurity capability is not, under this framework, a commercial transaction between a private company and its customers. It is a privilege whose extension is conditioned on a government-defined trust relationship — a relationship whose criteria are not published, whose determination process is not transparent, and whose application can be modified with the same speed and unilateral authority that characterised the original suspension.
This is, in structural terms, identical to the access control architecture that has governed sensitive defence technology transfer for decades. A foreign government seeking access to advanced American weapons systems does not purchase them on the open market. It applies through the Foreign Military Sales programme, undergoes an interagency review process, and receives access calibrated to the State Department’s and Defense Department’s assessment of the requesting government’s reliability, alignment with American strategic interests, and the technology’s sensitivity. The Mythos 5 episode demonstrates that an equivalent architecture is now operative for frontier AI capability — applied, in this first instance, primarily to domestic commercial access, but with an architecture that extends naturally, and that the export control directive’s foreign national provisions already partially extended, to international access.
Analyst note
The significance of demonstrating this capability domestically first should not be understated. The United States has effectively conducted a live-fire test of frontier AI access control — discovering, through the Mythos 5 episode, that tiered trust-based access restoration is achievable within two weeks once sufficient institutional motivation exists. This domestic demonstration provides the operational template for an international application: the same mechanism that restored Mythos 5 access to trusted US companies can, with comparatively modest adaptation, be extended to calibrate access for foreign governments, foreign companies, and foreign research institutions according to their alignment with American strategic objectives. The technology of the leverage has been proven. Its international application is the logical next phase.
AI Access as the New Currency of Alliance Management
The Pax Silica framework, documented earlier in this series, established critical minerals and semiconductor supply chains as the architecture through which the United States is organising a coalition of trusted technology partners against the alternative ecosystem China is constructing. Frontier AI access, on the evidence of the Mythos 5 episode and its surrounding context, is emerging as a parallel and potentially more powerful instrument in the same strategic architecture.
The reason frontier AI access functions as more potent leverage than mineral supply chains is structural: AI capability is not merely an input to economic production, as critical minerals are. It is increasingly the capability through which states conduct cyber defence, economic intelligence analysis, scientific research acceleration, and — as the cybersecurity-specific capabilities of models like Mythos 5 demonstrate — both offensive and defensive cyber operations themselves. A government granted privileged access to American frontier AI capability gains a tool that touches nearly every domain of state competence simultaneously. A government denied that access, or granted only degraded or delayed access relative to its competitors, experiences a comprehensive capability gap that no alternative single resource can fully compensate for.
The allied dimension of the Mythos 5 episode illustrates this dynamic directly, even though it has received less attention than the domestic regulatory dispute. Allied governments that had been granted access to Mythos 5 through the Project Glasswing cybersecurity defence programme since April lost that access without warning when the export control directive was issued, and the partial restoration of June 27 has not been confirmed to extend to those allied government users on the same terms, or any specified terms, as the “trusted” US companies and agencies whose access has been restored. An allied government’s cyber defence capability, calibrated around continued access to a specific American AI model, was demonstrated to be revocable at the discretion of the United States government with no advance consultation — a demonstration whose strategic message, intended or not, is unambiguous: American frontier AI capability sharing is conditional, can be withdrawn unilaterally, and allied governments’ defensive planning should account for that conditionality.
This is precisely the logic that has historically governed American intelligence sharing within the Five Eyes alliance and the broader network of bilateral intelligence relationships: access is calibrated to trust, trust is assessed continuously rather than granted permanently, and the withdrawal of access functions as both a punitive measure and a signal to other partners about the behaviour the United States expects in order to maintain access. Frontier AI capability sharing, on the evidence of the past month, is being absorbed into this same governance logic.
The Domestic Test Case as International Template
The decision to first exercise this leverage against a domestic American company — rather than against a foreign government or foreign AI developer — deserves analysis as a deliberate sequencing choice rather than merely the consequence of where the triggering vulnerability happened to be discovered.
Exercising new regulatory leverage against a domestic company carries lower diplomatic risk than exercising it against a foreign government, while still establishing the operational precedent, the legal authority’s practical scope, and the technical feasibility of the access control mechanism. The Mythos 5 episode allowed the United States government to discover, test, and refine the trusted-tier access framework in a context where the affected party — Anthropic — had strong commercial incentives to cooperate, no capacity to retaliate diplomatically, and no alternative government whose support it could invoke against the action. A foreign government subjected to an equivalent action, by contrast, would have diplomatic options, alliance relationships, and potential retaliatory measures available that a domestic company does not.
Having now demonstrated the mechanism domestically, the United States possesses both the proven technical architecture and the legal precedent — established through the Commerce Department’s export control authority — to apply equivalent access calibration internationally with considerably greater confidence in its practical workability than it would have possessed before the Mythos 5 episode. The parallel OpenAI development, in which GPT-5.6’s rollout was staggered at government request with early access limited to trusted partners during a security review, suggests this template is already being generalised across the domestic frontier AI industry before any international extension has been publicly confirmed — establishing, in effect, a domestic compliance norm that international partners and competitors will be evaluated against.
Analyst note
The choice of language in the OpenAI case — “trusted partners” receiving early access during a security review — uses identical framing to the Anthropic resolution, suggesting the Commerce Department or a coordinating body within the administration has developed, even if not yet published, a standardised vocabulary and process template for frontier model pre-release government engagement. A standardised process applied consistently across at least two major frontier AI developers within the same month is meaningfully different from an ad hoc, company-specific enforcement action. It indicates the early formation of an institutional doctrine — not yet codified in regulation or legislation, but operative in practice — for how the United States government intends to manage its relationship with frontier AI capability going forward, domestically and, by extension, internationally.
The China Contrast and the Competitive Logic
The strategic logic of developing AI access as an instrument of influence becomes clearest when considered against the China competition framework that has structured nearly every major US technology policy decision examined throughout this series.
China’s approach to AI capability as an instrument of international influence has followed a different model: the export of AI-enabled surveillance and governance technology to partner states, often bundled with infrastructure investment through Belt and Road-adjacent financing, creating dependency relationships calibrated around technology provision rather than technology restriction. The Digital Silk Road model extends Chinese AI and digital infrastructure capability outward, building influence through provision and integration.
The American model emerging from the Mythos 5 episode is structurally different: influence built through the control of access to a capability advantage rather than through the export of dependency-creating infrastructure. Where China’s model says “we will build your digital infrastructure and you will depend on our technology,” the American model implicit in the trusted-tier framework says “we possess capability you do not, and your access to it will be calibrated to your alignment with us.” Both models produce dependency and influence. They produce it through opposite mechanisms — provision-based influence versus access-restriction-based influence — and the choice between them reflects each country’s underlying strategic position: China, as the technology provider seeking market penetration, benefits from generous provision; the United States, as the capability leader seeking to preserve its advantage while extracting alignment, benefits from calibrated restriction.
This is the same logic that has historically distinguished American and Soviet, and later American and Chinese, approaches to military and intelligence technology sharing with partner states throughout the Cold War and its aftermath. The country with the capability advantage tends toward restriction-calibrated-to-trust; the country seeking to build influence through provision tends toward generous, dependency-creating export. Frontier AI access is now visibly following this established pattern.
What This Means for the Legislative and Multilateral Architecture
The emergence of AI access as an instrument of American statecraft has direct implications for the Great American Artificial Intelligence Act discussion draft and for the broader international AI governance architecture this series has documented throughout 2026.
A legislative framework that formalises mandatory risk assessment, IVO auditing, and critical safety incident reporting for frontier AI developers would, if enacted, create a parallel and potentially competing governance channel to the access-control-as-leverage architecture that the Mythos 5 episode has demonstrated. The legislative framework is oriented toward domestic safety assurance, applied consistently to all covered developers regardless of their international relationships. The access-control architecture that has emerged through executive action is oriented toward strategic calibration, applied selectively according to the government’s assessment of trust and alignment. These two governance logics are not inherently incompatible, but they answer different questions, and a frontier AI developer operating under both simultaneously will face a compliance environment that is, in important respects, less about safety in the abstract and more about demonstrating alignment with the strategic objectives the access-control architecture is designed to enforce.
The international governance implications are more significant still. The UN Global Mechanism’s norms-development process, and the broader multilateral architecture for responsible AI governance, presumes a degree of international cooperation and consistency in how frontier AI capability is governed. An American approach that treats frontier AI access explicitly as an instrument of bilateral influence — calibrated to alignment, withdrawable without consultation, and extended preferentially to trusted partners — sits uneasily alongside any multilateral framework premised on consistent, internationally coordinated governance standards. Allied governments and partner states will need to develop their own assessment of how much strategic dependency on American frontier AI capability they are willing to accept, given the demonstrated reality that such access is conditional and can be withdrawn with the same speed and limited consultation that characterised the original Mythos 5 suspension.
Bottom Line Assessment
The Mythos 5 episode, read in isolation, is a cybersecurity governance dispute resolved through negotiation. Read in the context this analysis has constructed, it is the operational debut of a new instrument of American international influence: calibrated, trust-conditioned access to frontier AI capability, exercised first domestically to establish the mechanism’s technical feasibility and legal authority, and now positioned for extension to the international partners and competitors whose strategic alignment with the United States the government has every incentive to want to influence.
This instrument shares the essential characteristics of every prior generation of American technology-based statecraft: it creates leverage through capability advantage rather than capability denial alone, it is exercised through processes whose criteria are not fully public, and it positions access itself — rather than the underlying technology in the abstract — as the unit of diplomatic and strategic currency. The parallel development at OpenAI suggests the mechanism is generalising rapidly across the American frontier AI industry, with or without formal codification in legislation or regulation.
The question this raises for the remainder of 2026 and beyond is not whether the United States will use frontier AI access as an instrument of influence — the Mythos 5 episode has already demonstrated that it will, and the technical and institutional mechanisms now exist. The question is how systematically this instrument will be formalised, how it will interact with the legislative governance framework moving through Congress, how allied governments will respond to the demonstrated conditionality of their own access, and how China’s alternative model of influence through generous provision will compete against an American model of influence through calibrated restriction.
The lever has been built and tested. Its deployment as a tool of international relations, rather than merely domestic regulation, is the development to watch in the months ahead.
AI Statecraft · Frontier AI Access · Anthropic · OpenAI · Export Controls · Technology Diplomacy · Pax Silica · US-China Competition · International Influence · Vladimir Tsakanyan
Leave a comment