BREAKDOWN : Trump, AI, and China — When the West's Survival Rests on a Flawed Executive Order

Introduction: America Bets Its Technological Dominance on an All-or-Nothing Gamble

An executive order signed in the shadows, away from cameras

On June 2, 2026, U.S. President Donald Trump quietly signed — without ceremony, without invited CEOs, without fanfare — an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security." No customary staging. Two weeks earlier, Trump had himself cancelled the signing of a more ambitious version of the same text, just hours before a planned Oval Office ceremony. His justification, blunt and revealing: he didn't want to do anything that might "get in the way of" America's competition with China for AI supremacy. That moment says everything about the fundamental tension running through this administration — and, more broadly, the entire West — as it confronts Beijing's technological rise.

Three days after that decree, on June 5, 2026, Trump signed National Security Presidential Memorandum NSPM-11, titled "Artificial Intelligence in the National Security Enterprise." This second document addresses the American military-industrial complex directly: it orders a complete overhaul of how the military, the intelligence community, and national security agencies adopt, adapt, and control AI. Together, these two documents form the most ambitious AI policy of Trump's second administration. And the most contested.

Two texts, one obsession: don't let China win

Behind the legal technicalities of these decrees lies a central obsession: the fear that China will surpass the United States in the race for artificial intelligence. This fear is not irrational — it is fed by concrete facts, documented intrusions, and Chinese technological performances that have surprised even the most confident Silicon Valley engineers. Early 2025 saw startup DeepSeek launch a model whose performance wiped hundreds of billions of dollars off the market capitalization of American tech giants. That was no accident — it was a strategic warning shot.

It is in this context that the two June 2026 texts must be read: not as ordinary bureaucratic regulations, but as foundational political acts of a new doctrine. A doctrine that acknowledges, for the first time explicitly in a text signed by Trump, that AI is no longer a matter of commercial innovation — it is a matter of national security, military power, and the strategic survival of the West in a world redefining itself at breakneck speed.

The June 2 Decree: What It Actually Says

A voluntary framework that stops halfway

The operational heart of the decree is what is called a voluntary pre-deployment framework: companies developing "covered frontier models" — that is, the most powerful AI models, those likely to have advanced cybernetic capabilities — are invited to voluntarily submit their systems to the U.S. government up to 30 days before public release. The key word is "voluntarily." The government cannot require this access. It requests it. That is a fundamental difference which, according to the text's critics, neutralizes much of its real-world reach.

The decree also orders the creation of an "AI cybersecurity clearinghouse" — involving the Treasury, NSA, and CISA — in voluntary cooperation with industry and critical infrastructure operators. It further orders the development of a classified benchmarking process to assess AI models' cybernetic capabilities, and extends access to cyber-defense tools to U.S. states, municipalities, and critical infrastructure operators — rural hospitals, community banks, local utilities. These are concrete, useful measures. But they are framed by a logic of cooperation, not compulsion.

What the decree explicitly prohibits: a guarantee for industry

A crucial provision of the text — often overlooked — specifies that nothing in the decree should be interpreted as authorizing the creation of a "mandatory governmental licensing, preclearance, or permitting requirement" for the development, release, or distribution of new AI models, including cutting-edge ones. In other words: the U.S. government ties its own hands. It may observe, test by invitation, advise — but it cannot block a public release. This is industry's political victory, inscribed in the very text of the decree.

This choice reflects the administration's economic philosophy: let market forces operate, avoiding what Trump himself called a "blocker." But in a race where China deploys its military AI models with direct state approval, without voluntary consultation procedures, this unilateral restraint raises a legitimate strategic question: can you win a race against an opponent running without the same constraints by deliberately refusing to use certain levers?

The 90-Day Version and the Capitulation of May 2026

A ceremony cancelled at the last minute

To understand what the June 2 decree truly represents, one must remember what it could have been. On May 21, 2026, Trump had convened a White House signing ceremony, with a list of Silicon Valley CEOs. The version then circulating provided for a government review window of up to 90 days before the most advanced models' public release. Hours before the ceremony, Trump cancelled. He told reporters, laconically: "I didn't like certain aspects of it. I postponed it." He explained that he feared certain provisions would become an obstacle at a moment when AI was generating investment and jobs.

According to Politico, AI industry officials had considered the 90-day window "too onerous." Some wanted it reduced to about two weeks. The version signed on June 2 gave them 30 days — a compromise that satisfies industry more than security experts. The fact that Trump signed the final text in private, having planned a grand public ceremony for the 90-day version, is an implicit admission: the administration knew this retreat would be hard to defend on camera.

Industry pressure as a major political force

The sequence from May 21 to June 2 reveals something structural in Trump's AI governance: industry is a de facto actor in the legislative process. The major Silicon Valley companies — which invested heavily in relationships with the administration, notably through donations to Trump's inaugural committees — managed within weeks to transform a 90-day requirement into a 30-day voluntary invitation. This is not corruption in the legal sense. It is something more insidious: a capillary connection between political and economic power that makes regulation structurally difficult to sustain against industry's preferences.

And this retreat has concrete consequences. According to TechCrunch, an earlier draft had provided a review window of 14 to 90 days, with assessment procedures developed by the Office of the National Cyber Director. The final version not only reduces the duration but simplifies assessment procedures and removes any mandatory enforcement mechanism. Result: a government that politely asks, and an industry that can politely say no.

NSPM-11: AI Enters the War Rooms

Four pillars of an unprecedented military doctrine

The NSPM-11 memorandum of June 5 is a document of a different ambition. It does not address the civilian industry: it addresses the Department of War — the Pentagon under its Trumpian name —, the Intelligence Community, and everything the text calls the "national security enterprise." Its structure rests on four pillars: accelerated adoption of AI, adaptation of commercial or open-source technologies for military needs, assurance of systems' reliability and controllability, and accountability in their use, to avoid abuses — censorship, ideological bias, illegal surveillance.

The memorandum states explicitly what it wants to achieve: maintaining "technical overmatch" against America's adversaries, and preserving a "decisive and enduring AI advantage." It replaces and revokes NSM-25 from the Biden era, which it describes as an outdated document that would have imposed excessive bureaucracy, fostered dangerous single-vendor dependencies, and made it difficult for the military to adopt the most advanced technologies. It is a prosecutorial indictment of the previous era serving to legitimize a radical strategic pivot.

Autonomous weapons systems: the most explosive question

One of NSPM-11's most consequential provisions is the instruction to the Secretary of Defense to submit a revised version of DoDD 3000.09 on autonomy in weapons systems within 90 days. This directive governs rules of engagement for machines potentially capable of triggering lethal strikes autonomously. Revising it in 90 days — a timeline military expert Jack Shanahan describes as tight, even dangerous for a substantive overhaul — represents a risky bet on the altar of speed.

The memorandum nonetheless insists that the chain of command must remain central to military decisions involving AI, and that systems must be "reliable, robust, steerable and controllable." It also requires, via contractual clauses, that no commercial entity be able to disable, degrade, or modify without authorization an AI system on which the military depends. This is a lesson drawn directly from the Anthropic conflict — but its contractual translation remains to be built, in a very tight timeframe.

The Anthropic Factor: When AI Meets Contract Law

A contractual dispute that reconfigured government policy

NSPM-11 names no company explicitly. Yet according to the specialized outlet Breaking Defense, it is clearly inspired by what experts call the "Anthropic debacle." The facts: Anthropic challenged the reported use of its model Claude to plan combat operations against Venezuela and Iran. In response, the Trump administration cancelled all federal contracts with Anthropic on the grounds that the company constituted a "supply chain risk." Anthropic responded with two parallel lawsuits. Meanwhile, one exception was preserved: the Claude Mythos model remains in use at the NSA, where it was the first large commercial language model authorized on classified networks.

NSPM-11 responds to this contractual chaos by establishing a general rule: any company whose conduct is incompatible with the memorandum's requirements may have its contracts terminated, including for "default or convenience." Exceptions are limited to a maximum of one year and must be reported in writing to the president's national security and technology advisers. Expert Mike Horowitz summarized it bluntly: "There is a lot more continuity here than change." In other words, the proclaimed break with Biden is less dramatic than announced. But the logic of subordinating industry to military priorities is clearly affirmed.

The danger of single-vendor dependency

Beyond the Anthropic case, NSPM-11 reveals a structural concern of the administration: dangerous dependence on a single vendor. The Claude affair demonstrated spectacularly the risk of entrusting the most critical military capability to a single private company that can, at any moment, contest the terms of its product's use. The memorandum's response is to diversify sources — by encouraging the adaptation of commercial or open-source technologies from multiple providers, large and small, and allowing agencies to develop their own in-house solutions when security requirements justify it.

This is a lesson Europe should heed. Dependence on American suppliers for critical technologies creates a strategic vulnerability analogous to what the United States discovered with its own private contractors. Diversifying AI sources for defense and national security is not a luxury — it is an existential necessity at a time when autonomous weapons systems are beginning to rely on language models for target recognition and real-time decision-making.

The Gutted CISA: Who Will Implement Any of This?

DOGE cuts and their consequences for cyber-defense

The June 2 decree assigns CISA (Cybersecurity and Infrastructure Security Agency) a central implementation role: publishing binding operational directives, expanding cybersecurity programs, building a classified AI evaluation system. The problem? CISA has lost about one-third of its workforce since the start of the Trump administration, according to Senator Mark Warner. Red teams responsible for simulating cyberattacks and identifying vulnerabilities were particularly hard hit. Acting director Nick Andersen was seeking to recruit some 300 critical personnel in March. Senator Markwayne Mullin estimated the agency needed approximately 2,800 agents to function properly.

The irony is cruel: at the very moment Trump signs an ambitious decree to strengthen America's cyber-defense through AI, the agency supposed to lead its implementation is hollowed out. In August 2025, CISA itself had published, jointly with the NSA and FBI, an urgent advisory warning that Chinese state-linked hackers associated with the Salt Typhoon campaign had infiltrated American telecommunications networks, exposing sensitive data on presidential candidates and members of Congress.

The Trumpian paradox: building on gutted foundations

Expert Richard Forno of the University of Maryland said it plainly: the way the federal government has treated its employees over the past year — through OPM and DOGE — makes it very unlikely that future cybersecurity professionals will choose federal service. Recruiting cybersecurity experts is already difficult in the private sector, where salaries are significantly higher. After a wave of layoffs that hit entire teams, including election security specialists, the appeal of public service in this domain has diminished further.

This paradox — ambitious on objectives, destructive on means — is not an isolated miscalculation. It reflects a coherent worldview: the conviction that government is inherently inefficient, and that the solution to every problem is to reduce the size of the state and entrust critical functions to private actors. Applied to national cybersecurity, this philosophy produces predictably problematic results: a weakened agency, exposed networks, and a decree assigning that same agency new responsibilities without giving it the resources to assume them.

China as Catalyst: DeepSeek, Salt Typhoon, and Strategic Fear

A technological adversary that closed the gap faster than expected

Behind all these texts, there is an engine: fear of China. Not an abstract fear — a fear nourished by documented facts. In early 2025, Chinese startup DeepSeek launched its flagship V3 model, whose performance shook Silicon Valley's confidence and wiped hundreds of billions of dollars off the market capitalization of major American technology companies. This seismic event proved that China is no longer merely catching up to the United States in AI — it is challenging them on efficiency and performance-to-cost ratios. In parallel, the Salt Typhoon espionage campaign exposed the deep vulnerability of America's telecommunications infrastructure to Chinese state actors.

According to The Diplomat, advanced AI models are now treated in Washington as strategic assets on par with semiconductors, telecommunications, and critical minerals. The stakes are no longer just about building the best models — it is about deciding who sets the standards, who is deemed trustworthy, who gains access to strategic markets. The fundamental question posed by Trump's order is simple: can the West maintain its technological edge while refusing to put in place the safeguards that would prevent catastrophe in an unchecked race?

The Chinese doctrine: regulate and innovate simultaneously

A fact often ignored in American debates on AI regulation deserves highlighting: since 2021, China has put in place the world's most extensive and constraining AI regulations. Rules on recommendation algorithms, deepfakes, and since 2023 specific regulations on generative AI. These regulations require pre-registration of models, rigorous testing including thousands of questions on politically sensitive topics, and the obligation to obtain compliant responses to at least 90% of these questions before public release.

Result? Despite these constraints — or perhaps because of them — China has produced competitive models and has even outpaced its American peers on efficiency metrics. According to analyst Matt Sheehan of the Carnegie Endowment for International Peace, Chinese companies have produced models that rival their American competitors in capabilities, while complying with a highly constraining regulatory framework. Reducing the AI race to a binary choice between regulation and innovation is therefore a false dichotomy — and a particularly dangerous false dichotomy when it serves to justify the absence of safeguards.

The Carnegie Argument: Regulating Is Not Losing

Dismantling the anti-regulation rhetoric

The analysis published by the Carnegie Endowment for International Peace in early June 2026 directly challenges the central argument of the Trump administration for weakening safeguards. Matt Sheehan demonstrates that the opposition between regulation and AI competitiveness is built on a factual error: China has proven for four years that one can regulate seriously and innovate massively. The true determining factors in the AI race are not regulatory rules — they are access to capital, research talent, and computing power. On all three counts, the United States still holds a considerable lead over China.

The article also notes that China's AI governance system is not heavy European-style bureaucracy — it is a small, fast, and flexible system. China has published targeted regulations on specific problems, iteratively, leaving technical details to standards developed in direct collaboration with companies, researchers, and officials. This model has made Chinese regulators technically competent and close to the ground — the opposite of what American laissez-faire advocates fear.

What the United States could learn

Matt Sheehan points to the Center for AI Standards and Innovation at the Commerce Department as an example of what works: technically competent agents working directly with OpenAI and Anthropic, conducting sophisticated tests to detect security risks. This is precisely the kind of institution that the Trump administration's budget cuts and anti-regulation philosophy endanger. The right response to the Chinese threat is not to destroy regulatory agencies — it is to make them more competent, faster, and closer to the technology.

This is a lesson that also applies to Europe. The European AI Act, with its complex compliance requirements and implementation timelines stretched over several years, represents the exact opposite of the Chinese model — and a real risk to the competitiveness of European companies. The challenge for the West is to find the middle path: governance that protects without blocking, that controls without stifling. That narrow road is the only one that allows remaining in the race while staying faithful to the values that make our democracies strong.

The Diplomat: Toward a World AI Order Divided into Two Spheres

The balkanization of technological ecosystems

Analyst Guangyi Pan in The Diplomat brings a decisive geopolitical perspective: Trump's order accelerates a dynamic already underway — the construction of two parallel technological spheres. On one side, a trusted ecosystem for American and allied models, structured around defense partnerships and intelligence-sharing agreements. On the other, a separate circuit for Chinese models, subject to structural suspicion over questions of backdoors, data transfer to Beijing, and alignment with Chinese Communist Party objectives.

Chinese AI companies have already penetrated parts of the American digital ecosystem through open-source models, developer platforms, and consumer applications. But they will likely never be "trusted partners" for American cyber-defense or critical infrastructure. This is a strategically correct conclusion, even if it creates a global technological fracture whose consequences — notably for Global South countries that will have to choose sides — are not yet fully measured.

The risk of concentration benefiting tech giants

What Pan raises with precision is the way the American voluntary system could become, in practice, a de facto obligation. If investors, institutional clients, and government agencies treat participation in the pre-deployment program as a trust signal, companies that refuse to participate will be commercially stigmatized. Large companies like OpenAI, Google, and Anthropic are better equipped to navigate this relationship with the state. Smaller companies and open-source developers risk being structurally disadvantaged.

This is a form of regulation that dare not speak its name — and that concentrates power in the hands of a handful of already dominant players. The long-term consequence evoked by The Diplomat is a more divided global AI order, in which technical standards, security standards, and trust criteria will be defined by the United States and its allies on one side, and by China and its sphere of influence on the other. This is not a hypothetical dystopia — it is a trend already observable in the semiconductor and telecommunications domains.

Classified Benchmarking: Governing in the Shadows

An evaluation mechanism out of public sight

One of the decree's most concerning provisions — and the least discussed in mainstream media coverage — is the instruction to develop and maintain a classified benchmarking process to assess the advanced cybernetic capabilities of AI models. The analysis by law firm JD Supra highlights this explicitly: at least part of the administration's evaluation framework will operate out of public view. Concretely, the criteria for determining that a model is powerful enough to trigger a government review will not be public. Nor will the results of these evaluations. Nor the resulting decisions.

This logic of secrecy — defensible in certain national security contexts (one does not publicly publish methods for testing whether an AI can compromise critical infrastructure) — creates a fundamental governance problem. How can companies comply with rules whose precise criteria they do not know? How can independent researchers, legislators, and civil society assess whether this system is working properly?

The bipartisan signal of an acknowledged insufficiency

The bipartisan Congressional reaction, noted by JD Supra, suggests that the current framework "may be only a first step" rather than the endpoint of federal AI oversight. In other words: even Trump's political allies privately acknowledge that this decree will not be the last word on AI governance. The pressure for more constraining mechanisms will come — probably after the first major incident involving an advanced AI model in a cyberattack or a catastrophic military decision. This is the traditional mode of American regulation: act after the disaster, not before.

The binding operational directives the Secretary of Homeland Security must publish within 30 days will be critical in determining whether this framework has real reach. These directives will cover three priority areas: accelerating federal cyber-defense, expanding AI-assisted cybersecurity programs, and facilitating federal agencies' access to cutting-edge models. The quality of these directives — and the resources allocated to their implementation — will say more about the seriousness of the undertaking than the text of the decree itself.

Trump vs. Biden: Two Philosophies, One Challenge

The philosophical break behind the decrees

To grasp what these June 2026 texts represent, they must be recontextualized within the break Trump made on his very first day back in the Oval Office in January 2025: the abrogation by decree of Biden's 2023 AI Executive Order. That Biden-era text organized AI governance around a broad spectrum of risks: privacy, discrimination, disinformation, consumer protection, model safety. Trump replaced this architecture with a radically different vision: AI is a tool of national power, and the priority is innovation, cybersecurity, and geopolitical competition — not protecting individuals from bias or problematic uses.

This is a clear philosophical choice, and it is not without logic. In a world where China deploys AI in its military and surveillance systems with little regard for human oversight or civil liberties — per the very terms of NSPM-11 — it is legitimate to ask whether the West can afford the luxury of AI governance designed primarily to protect chatbot users from discrimination. The question is real and deserves to be asked.

The wrong answer to the right question

But Trump's response — eliminating safeguards rather than adapting them — is the wrong answer to the right question. The strength of the West has always been demonstrating that one can be both powerful and responsible. The argument that the pressure of Chinese competition compels us to abandon our democratic standards of technology governance is precisely the type of argument Beijing hopes to see prevail in Washington. If the West begins to imitate its adversaries' governance methods to defeat them, it validates their model and weakens its own legitimacy as a superior alternative.

A middle path exists — and there are precedents. Semiconductor export controls, the Commerce Department's entity list, restrictions on investments in strategic Chinese technologies: these mechanisms combine geopolitical pragmatism and procedural rigor. They show it is possible to protect Western advantage without renouncing democratic decision-making procedures. This model — rigorous, targeted, transparent — should inspire AI governance, not the voluntary decree signed in private.

The White House Fact Sheet: The Politics of Narrative

What the administration wants us to remember

The official Fact Sheet published on June 2 by the White House summarizes in political language the strategic vision carried by these two texts. The title sets the tone: "President Donald J. Trump Promotes Advanced Artificial Intelligence Innovation and Security." The keywords: innovation, security, global leadership, American supremacy. The key phrase from Trump's text: "We will continue to lead an America First cybersecurity effort that enhances both our national security and our global AI dominance." Global dominance — the word is dropped, unabashedly.

This political communication has a dual objective. First, reassure industry: the government will not be a gatekeeper, will not create a mandatory licensing system, will not slow innovation. Second, project strength toward China and the rest of the world: the United States has a strategy, is serious, and will not let its adversaries seize the technological advantage. This is a coherent message on the communications plane. The problem is that the reality of implementation — a weakened CISA, a voluntary framework without enforcement mechanisms, an opaque classified benchmarking process — does not match the stated ambition.

The gap between words and actions

A presidential Fact Sheet is by definition a political marketing document. But the gap between the grandiloquence of the language and the modesty of the mechanisms put in place deserves to be measured. "Global AI dominance" — and to achieve it, companies are politely asked to kindly show the government their models thirty days before releasing them. This gap is not merely a communications issue: it reflects the real tension between the political ambition to appear strong and the philosophical reluctance to effectively regulate the market.

The second national security presidential memorandum of June 12, 2026 — the Fact Sheet on protecting warfighters and intelligence officers against cyber threats — extended this framework with additional measures to secure military systems. But here too, the question arises: who will implement these measures, with what experts, and within what timelines? The coherence between the discourse of power and operational realities remains this administration's Achilles heel on the AI file.

What This Means for the Future of the Technological West

The urgency of a common doctrine

The question posed by these two June 2026 texts is, at its core, the same as that posed by all the great technological tipping moments of contemporary history: can one remain ahead in the race without sacrificing what makes the race worthwhile? The AI race between the United States and China is real, deep, and engages the future of democratic societies. Beijing is investing massively, recruiting massively, and building military and civilian AI systems without the ethical and legal constraints that democratic societies impose on themselves. The West's technological advantage — a real but fragile advantage — is what guarantees that open, pluralistic, and free values will continue to structure the international order.

Trump's decrees are not the ideal response to this challenge. But they testify to an awareness — belated, imperfect, distorted by industry interests — that AI is no longer a technological topic. It is a matter of national security, geopolitical power, and civilizational architecture. Europe, which chose the heavy regulatory path with the AI Act, risks missing this competition through excessive caution. The United States risks winning the deployment race while creating strategic vulnerabilities through excessive imprudence. The narrow path combines speed, ambition, and intelligent safeguards.

The price of negligence in the race that matters

The West has already missed several appointments with the consequences of its technological recklessness: dependence on Chinese digital platforms, disorganization in the face of electoral deepfakes, the absence of common norms on the autonomy of combat drones. Each time, the reaction came after the fact, in emergency, without strategic coherence. AI governance risks reproducing this pattern at an incomparably larger scale, with incomparably more powerful technologies. The June 2026 texts could be the beginning of a more serious doctrine — or another missed appointment dressed up as progress. The difference will be decided by the quality of implementation, the robustness of institutions, and the political will to hold the course when industry pushes again to soften the constraints.

What is certain is that inaction is no longer an option. The Salt Typhoon campaign, DeepSeek, the Anthropic litigation, the revision of DoDD 3000.09: all these signals converge toward the same conclusion. AI is already at the heart of strategic competition between the great powers. The question is no longer whether the West must engage — it is already engaged. The question is whether it engages with the rigor, coherence, and resources the stakes demand. For now, the answer is: not yet.

Conclusion: A Necessary Evil That Must Learn Not to Be Negligent

Urgency does not justify recklessness

The June 2026 texts are, in their way, a moment of truth for American technology policy. They establish that AI is a matter of state — a matter of strategic survival, even. They break with the initial laissez-faire of the first phase of Trump's second administration, which had revoked Biden's safeguards without putting anything in their place. And they lay the groundwork for a dialogue — admittedly imperfect and voluntary — between the state and the companies developing the most powerful technologies in human history. That is real progress, and it should not be denied.

The demand for a West that wins and remains itself

But real progress on a weakened foundation remains insufficient. The West — and the United States first of all — does not have the luxury of choosing between speed and responsibility. It needs both, simultaneously, because this is precisely what differentiates it from its adversaries. The AI race against China is one that must be won. But winning it by emptying regulatory agencies of their experts, by signing decrees behind closed doors, and by substituting wishful thinking for the binding framework the situation demands — that is running toward victory while shooting oneself in the foot. The West can do better. It must do better. And Trump's decrees, despite their flaws, at least remind us that the stakes are existential and that inaction is the one luxury we can no longer truly afford.

Sources

Primary Sources

White House — Executive Order "Promoting Advanced Artificial Intelligence Innovation and Security" — June 2, 2026

White House — National Security Presidential Memorandum NSPM-11 "Artificial Intelligence in the National Security Enterprise" — June 5, 2026

White House — Official Fact Sheet on the AI and Security Executive Order — June 2, 2026

Secondary Sources

Politico — "Trump signs downsized AI order" — June 2, 2026

TechCrunch — "Trump signs narrower executive order on AI oversight after industry objections" — June 2, 2026

Broadband Breakfast — "Trump Signs Downsized AI Cybersecurity Order" — June 2, 2026

JD Supra — "White House Issues Executive Order on AI Innovation and Security" — June 4, 2026

Broadband Breakfast — "One Year After DOGE Cuts, Cybersecurity Agency Struggles Over Staffing" — June 13, 2026