On a Friday morning, the Bureau of Industry and Security published 139 pages of new export control regulations. The rules prohibited US companies from selling advanced semiconductors — chips with more than 3.5 trillion operations per second — to China without a license that would almost never be granted. They also barred US persons from supporting Chinese advanced-chip facilities. Nvidia's stock dropped 12 percent before lunch. But the rules were not about stock prices. They were about who would control the compute infrastructure underlying every future AI system on earth.
The competition between the United States and China over artificial intelligence is frequently described as a "race," but the metaphor is imprecise. A race has a single finish line. The US–China contest is better understood as a multi-domain struggle for position across semiconductors, data, talent, standards, and strategic applications — each domain with its own dynamics and its own flashpoints.
The US retained dominant advantages through 2022 in chip design (NVIDIA, AMD, Intel), chip manufacturing equipment (Applied Materials, Lam Research, KLA), chip fabrication (TSMC in Taiwan, Samsung in South Korea — both reliant on US-origin technology), and frontier AI research (Google DeepMind, OpenAI, Anthropic). China led in data volume (1.4 billion citizens, comprehensive surveillance infrastructure), deployment speed in certain industrial applications, and total number of AI patent filings — though patent volume imperfectly measures capability.
The October 2022 controls represented a deliberate US decision to weaponize its chokehold on semiconductor supply chains. The key technical thresholds targeted chips capable of training frontier AI models: Nvidia's A100 and H100, as well as future equivalents. The rules were updated in October 2023 to close workarounds through which Nvidia had designed slightly less powerful chips (A800, H800) that nominally fell below the original thresholds.
On August 29, 2023 — the day US Commerce Secretary Gina Raimondo landed in Beijing — Huawei quietly began selling the Mate 60 Pro, a smartphone powered by the Kirin 9000s chip fabricated by SMIC at a 7nm process node. This was widely regarded as impossible given US export restrictions on advanced chipmaking equipment. Subsequent teardowns confirmed 7nm production, demonstrating that restrictions created friction and delay but had not achieved technological containment. The chip showed yields and performance inferior to cutting-edge TSMC nodes, but the symbolic and strategic significance was enormous.
The semiconductor supply chain is extraordinarily concentrated at several chokepoints that the United States and its allies have leveraged. ASML, a Dutch company, is the sole manufacturer of extreme ultraviolet (EUV) lithography machines, which are required to produce chips at sub-7nm nodes. Under US pressure, the Netherlands restricted ASML from exporting EUV machines to China starting in 2019. Without EUV access, Chinese foundries cannot replicate leading-edge fabrication processes used by TSMC or Samsung.
The CHIPS and Science Act, signed by President Biden on August 9, 2022, allocated $52.7 billion for domestic semiconductor manufacturing and research, with a "guardrail" provision prohibiting recipients from expanding advanced chip capacity in China for ten years. Intel, TSMC, Samsung, and Micron subsequently announced or accelerated US fab projects totaling over $200 billion in combined investment.
China responded through the "Big Fund" (National Integrated Circuit Industry Investment Fund), which by 2024 had deployed over 340 billion yuan (~$47 billion) into domestic chipmakers including SMIC, Yangtze Memory Technologies (YMTC), and CXMT. YMTC had achieved competitive 3D NAND flash memory before being added to the US Entity List in December 2022.
The AI talent dimension of the competition is complex and defies simple nationalist framing. A 2023 MacroPolo study found that 38 percent of top-tier AI researchers working in the United States were born in China — making Chinese-born researchers the single largest non-American cohort in US AI. Simultaneously, roughly 25 percent of top AI researchers at Chinese institutions received their doctoral training in the United States.
US visa policy has oscillated between viewing Chinese AI researchers as a talent asset and a security risk. The Trump administration suspended certain exchange-visitor visas for Chinese nationals affiliated with institutions linked to the PLA. The Biden administration maintained scrutiny while attempting to distinguish between legitimate academic exchange and technology transfer tied to military end-users.
China's domestic training pipeline has expanded dramatically. The 2017 New Generation AI Development Plan set a target of becoming the world's leading AI innovation center by 2030. By 2023, Chinese universities were producing more STEM PhD graduates annually than the United States, though research citation impact and top-tier conference publication rates still favored the US.
The consensus among US national security analysts as of 2024 was that China lagged approximately one to two years behind the US frontier in large language models and advanced AI systems — a gap maintained largely by semiconductor restrictions — but was closing rapidly in applications including computer vision, autonomous systems, and AI-enabled surveillance. Whether restrictions could sustain a meaningful capability gap beyond a five-year horizon was actively debated.
You are advising a Senate committee on semiconductor export policy. The AI assistant will help you think through the strategic logic, tradeoffs, and second-order effects of the October 2022 BIS controls and their successors.
Engage with at least three substantive exchanges to complete this lab.
Ukrainian forces reported that Russian Lancet-3 loitering munitions — small, cheap, and guided by computer vision — had destroyed a HIMARS launcher at a cost of roughly $35,000 per weapon versus a system valued at over $5 million. The Lancet used optical target recognition trained on images of NATO equipment. It was, in miniature, a preview of what AI-enabled attrition warfare looks like: the ability to destroy expensive precision assets with cheap, autonomous, mass-produced weapons.
The conflict in Ukraine has served as a live testing ground for AI-enabled military systems on both sides. Russia's deployment of AI has been constrained by sanctions cutting off access to Western semiconductors, but several documented systems have demonstrated meaningful AI-enabled capabilities.
The Lancet-3 loitering munition, produced by ZALA Aero (a Kalashnikov subsidiary), uses computer vision for terminal guidance. It achieved operational effectiveness against Ukrainian artillery, armored vehicles, and air defense systems throughout 2022–2024. Ukrainian military intelligence documented over 200 confirmed kills attributed to Lancet strikes on high-value systems.
Russia has also deployed Orion UAVs with AI-assisted targeting and electronic warfare systems using machine learning to identify and jam Ukrainian communication frequencies. The Shahed-136 (Iranian-designed, Russian-deployed) one-way attack drone uses inertial navigation and terrain-matching algorithms rather than true machine learning, but demonstrated the value of cheap autonomous munitions at scale — with over 2,000 launched against Ukrainian infrastructure by early 2024.
Sanctions have imposed real costs. Analysis of captured Russian electronics revealed continued reliance on foreign chips — particularly from Texas Instruments, Intel, and Analog Devices — often acquired through third-country intermediaries in Turkey, UAE, and Central Asia. A February 2023 Royal United Services Institute (RUSI) report documented over 450 Western-origin components in captured Russian precision munitions.
Ukraine has deployed AI tools more extensively and creatively than Russia in several domains. Palantir's AI-enabled targeting platform, deployed to Ukrainian forces in 2023, integrates satellite imagery, drone feeds, and battlefield sensor data to identify Russian positions and recommend fire missions. Ukraine's Delta battlefield management system uses AI to fuse intelligence from multiple sources into a common operational picture. Ukraine also deployed facial recognition (Clearview AI) at checkpoints and to identify Russian soldiers killed in action — raising significant IHL questions.
Iran's Shahed series — particularly the Shahed-136 Geran-2 as renamed for Russian use — represents a deliberate strategy of using AI-adjacent technologies (terrain navigation, target recognition) to produce highly cost-effective attrition weapons. Iran transferred hundreds of Shahed-136s to Russia in 2022 and accelerated production through a joint factory established in Alabuga, Russia (confirmed by US intelligence in 2023).
Iran's domestic drone program has expanded to include the Mohajer-6, capable of carrying Qaem precision-guided munitions with electro-optical seekers that use image correlation for terminal guidance — a form of AI-assisted targeting. The Mohajer-6 has been documented in use by Hezbollah (supplied by Iran) for ISR missions over Israel and northern Israel surveillance since 2021.
Iran has also acquired Chinese AI surveillance technology — including Hikvision cameras and Huawei network infrastructure — for domestic repression. During the 2022 Mahsa Amini protests, Iranian security forces used AI-enabled facial recognition to identify and subsequently arrest demonstrators, according to Amnesty International documentation.
Both Russia and Iran have demonstrated sophisticated capabilities to evade technology export controls through layered procurement networks. Key documented methods include: front companies in neutral jurisdictions (UAE, Turkey, Armenia, Georgia) that purchase chips and electronic components and re-export them without disclosure; academic and research institution covers used to acquire dual-use computing equipment; and gray market acquisition of used enterprise servers and GPU clusters.
The US Treasury and Commerce Departments have responded with expanded designations. In June 2023, Treasury sanctioned over 70 entities in multiple countries for facilitating Russian technology procurement. The Justice Department established a "KleptoCapture" task force specifically focused on technology sanctions evasion. Despite these efforts, RUSI estimated in 2024 that Russia was successfully acquiring roughly 70–80% of its pre-war baseline semiconductor supply through alternative channels.
The Ukraine conflict has demonstrated a consistent pattern: AI is not yet making strategic decisions, but it is dramatically compressing the sensor-to-shooter loop at the tactical level. Systems that previously required minutes to process ISR data and generate targeting solutions now operate in seconds. This compression favors offensive operations and creates new vulnerabilities for static defensive positions — a dynamic that AI researchers and military planners describe as a fundamental shift in combined arms warfare.
You are a defense analyst preparing a classified assessment of AI's documented operational role in the Ukraine conflict. The AI assistant will help you analyze specific systems, their documented effects, and the implications for US military planning.
Engage with at least three substantive exchanges to complete this lab.
Huawei built Kenya's National Integrated Identity Management System (NIIMS) — a biometric database containing fingerprints, facial recognition data, and iris scans for every Kenyan citizen. The Huduma Namba card became the gateway to government services. Critics noted that the system's data architecture gave Huawei access to the database under maintenance contracts, and that the system's design reflected surveillance norms — mandatory enrollment, centralized storage, no independent oversight mechanism — that had been built into Chinese domestic systems first.
China's Digital Silk Road (DSR), formally part of the Belt and Road Initiative since 2015, involves the export of Chinese telecommunications infrastructure, AI surveillance systems, data centers, and smart city platforms to developing and middle-income countries. By 2023, DSR projects had been documented in over 80 countries across Africa, Southeast Asia, South Asia, the Middle East, and Latin America.
The major Chinese technology companies involved include Huawei (telecommunications, safe cities, cloud), ZTE (telecommunications), Alibaba Cloud (data centers, e-commerce platforms), Hikvision and Dahua (AI surveillance cameras), SenseTime and Megvii (facial recognition software), and CloudWalk (biometric databases). Many of these companies are on US government restricted lists but continue to operate freely in markets not subject to US jurisdiction.
Safe City programs are the most documented DSR AI application. These integrated platforms combine CCTV networks (often with AI-enabled facial recognition), traffic monitoring, gunshot detection, social media monitoring, and sometimes predictive policing algorithms. Huawei's Safe City platform had been deployed in over 700 cities across more than 100 countries as of 2022, according to company marketing materials and independent research by Steven Feldstein at the Carnegie Endowment.
Ecuador's ECU-911 system, built by CICC (a Chinese state-linked company) with significant Huawei components, integrated over 4,300 cameras in 24 cities, connecting police, fire, and emergency services through a centralized command center. The system was praised for reducing response times and cutting crime statistics. However, an investigation by the New York Times in 2019 found that Ecuadorian security services had used the system for political surveillance, and that China retained technical access to the camera network under maintenance agreements. Ecuador subsequently sought to replace Huawei components following US diplomatic pressure in 2020–2021.
A growing body of academic and policy research argues that Chinese AI exports are not merely commercial transactions but vectors for the export of a specific governance model. The key argument is that technical choices embed normative assumptions: centralized data storage vs. distributed; mandatory enrollment vs. opt-in; law enforcement access vs. warrant requirements; cross-modal data fusion vs. siloed systems.
Samantha Bradshaw and colleagues at the Oxford Internet Institute documented in a 2021 study that countries using Chinese AI surveillance infrastructure were statistically more likely to increase authoritarian governance indicators in the five years following deployment, though the causal direction remains contested — authoritarian governments may preferentially select Chinese systems rather than Chinese systems causing authoritarian tendencies.
The Beijing AI Principles (2019) and China's domestic AI governance documents emphasize state oversight, social harmony, and the right of governments to determine the parameters of AI deployment on their territory — values that diverge significantly from EU-style rights-based frameworks and US emphasis on individual liberty.
The US government response to Chinese DSR AI has included several documented interventions. The Clean Network initiative (2020) sought to persuade allied governments to exclude Huawei from 5G infrastructure, with mixed success — the UK banned Huawei from 5G core networks in July 2020; Germany resisted until 2024. The US International Development Finance Corporation (DFC) and USAID have funded alternative digital infrastructure projects, including the Partnership for Global Infrastructure and Investment (PGII), announced at the 2022 G7 summit.
Japan's Free and Open Indo-Pacific strategy includes a digital component emphasizing "data free flow with trust" — contrasting with what Japanese officials characterize as Chinese data sovereignty models. The Quad (US, Japan, Australia, India) established a technology working group in 2021 focused partly on countering DSR influence in the Indo-Pacific.
However, competitive responses have been limited by the fact that no Western company offers an equivalent to Huawei's integrated, affordable, all-in-one smart city platform. Ericsson and Nokia provide telecommunications equipment but not the full-stack AI surveillance and smart city integration that Huawei bundles at subsidized prices. This creates a structural advantage for Chinese DSR that diplomatic pressure alone has not overcome.
By 2024, Chinese AI surveillance technology was embedded in the physical infrastructure of dozens of governments that the US considers strategic partners or swing states. Removing it carries transition costs measured in billions of dollars and years of disruption. The DSR has created durable dependencies that outlast any single political decision — a form of technological statecraft that operates on a timescale of decades rather than electoral cycles.
You are a State Department policy planner developing a strategy to counter Chinese DSR influence in a swing-state partner country that has already deployed Huawei Safe City infrastructure. The AI assistant will help you think through options, constraints, and tradeoffs.
Engage with at least three substantive exchanges to complete this lab.
The United Kingdom hosted the first global AI Safety Summit at Bletchley Park — the site where Alan Turing's team broke the Enigma code. Twenty-eight governments signed the Bletchley Declaration, acknowledging that AI posed potential catastrophic risks and committing to information sharing on frontier model safety. Notably, China signed. The US and EU signed. But the declaration was non-binding, contained no enforcement mechanism, and defined "frontier AI" narrowly enough that most military AI systems fell outside its scope. The summit was simultaneously a genuine diplomatic achievement — the first multilateral AI safety agreement — and a demonstration of how far any binding governance framework remained.
Technical standards for AI — which metrics define safety, which architectures are permissible in critical infrastructure, which evaluation benchmarks determine regulatory compliance — are being set in international bodies where the US and China are in active competition. The key battlegrounds include the International Organization for Standardization (ISO), the International Electrotechnical Commission (IEC), and the International Telecommunication Union (ITU).
China has pursued an explicit "standards first" strategy in AI governance. A 2019 analysis by the Information Technology and Innovation Foundation (ITIF) found that Chinese entities held leadership positions in a disproportionate share of ITU AI working groups relative to their technical contributions. China has proposed and advanced standards emphasizing state authority over AI systems, mandatory government backdoor access to AI platforms, and social credit-compatible behavioral norms — proposals that have been contested by US, EU, and Japanese delegations.
The US National Institute of Standards and Technology (NIST) AI Risk Management Framework (published January 2023) represents the US approach: voluntary, risk-based, flexible, developed through industry consultation. The EU AI Act (formally adopted June 2024) takes a harder regulatory approach with mandatory requirements for high-risk systems and prohibited uses. Neither framework has achieved global adoption, and the divergence between them complicates allied coordination.
At the May 2023 G7 Summit in Hiroshima, leaders launched the "Hiroshima AI Process" to develop international guiding principles and a code of conduct for advanced AI. The process produced an 11-point code of conduct in October 2023 emphasizing transparency, safety testing, and incident reporting for frontier AI developers. Like the Bletchley Declaration, it was non-binding. But it established a Western-led multilateral forum for AI governance operating in parallel to — and often in contrast with — Chinese-led ITU processes.
The sharpest divergence in AI governance norms concerns autonomous weapons systems. The United States formally adopted a Department of Defense Directive 3000.09 (original 2012, updated 2023) requiring "appropriate levels of human judgment" over lethal force. The directive does not mandate human-in-the-loop for all weapons but requires that systems be designed, tested, and deployed in ways that allow commanders to exercise adequate control.
China has not adopted an equivalent policy and has resisted international negotiations on lethal autonomous weapons systems (LAWS) at the UN Convention on Certain Conventional Weapons (CCW). At CCW meetings from 2017–2024, China consistently opposed legally binding instruments while nominally supporting "political declarations." Russia similarly opposed binding restrictions.
The US and 57 allies signed a Political Declaration on Responsible Military Use of Artificial Intelligence at a State Department event in February 2023 — the largest multilateral commitment on military AI to date. Notably absent: China, Russia, Iran, and several other states whose AI programs are most relevant to the risks the declaration sought to address.
A separate bilateral channel exists: the US–China military-to-military relationship has included limited discussions on AI risk reduction, particularly on crisis communications and hotline protocols to prevent AI-related misunderstandings from escalating. The November 2023 Biden–Xi Sunnylands summit included agreement to continue AI safety dialogue — a modest but symbolically significant step.
Beyond standards, a new governance debate has emerged around compute governance — the idea that restricting access to the physical computing infrastructure needed to train frontier AI models could serve as a mechanism for AI safety oversight. The core argument, developed by researchers at the Center for Security and Emerging Technology (CSET) and the Future of Life Institute, is that because advanced AI training requires large clusters of specialized chips (currently mostly Nvidia GPUs), export controls on those chips could function as a de facto global frontier AI governance mechanism.
This logic underlies the October 2022 BIS controls: limiting China's access to A100/H100 chips limits China's ability to train frontier models that could be applied to military purposes. But the governance-through-compute approach raises profound equity questions: it concentrates frontier AI development in a small number of wealthy, primarily Western countries and requires accepting that a commercial firm (Nvidia) and a regulatory agency (BIS) are jointly determining the global distribution of transformative AI capability.
The International AI Safety Report, commissioned by the Bletchley summit and produced by a panel of 30 experts led by Yoshua Bengio, published in January 2024, explicitly recommended that governments consider compute monitoring as a governance tool — the first major international AI safety document to do so.
AI geopolitics in the governance dimension reflects a fundamental tension: effective AI safety governance requires international cooperation, including with China and Russia. But those same states are using international governance processes to advance standards and norms that undermine the individual rights frameworks and democratic oversight mechanisms that Western governments seek to embed in AI development. The governance competition is, at its core, a competition over which values will be institutionalized in the infrastructure of the 21st century.
You are advising the National Security Council on US strategy for the next major international AI governance negotiation. The AI assistant will help you evaluate different governance approaches, analyze precedents from other technology governance regimes, and stress-test proposed mechanisms.
Engage with at least three substantive exchanges to complete this lab.