U.S. State Department orders global warning about alleged AI thefts by DeepSeek, other Chinese firms

The United States State Department issued a diplomatic cable on 24 April 2026 directing all diplomatic and consular posts worldwide to alert foreign counterp...

What Happened

The United States State Department issued a diplomatic cable on 24 April 2026 directing all diplomatic and consular posts worldwide to alert foreign counterparts about alleged intellectual property theft by Chinese artificial intelligence companies, specifically naming DeepSeek, Moonshot AI, and MiniMax.
The cable, characterised as a global warning, instructed diplomatic staff to raise "concerns over adversaries' extraction and distillation of US AI models" with their host governments.
OpenAI had earlier (February 2026) formally alleged in a memo to the US Congress's China Select Committee that DeepSeek had used "extraction and distillation" techniques — exploiting access to ChatGPT systems through third-party routers — to train its own models on proprietary data.
The Chinese government denied the allegations as "groundless" and characterised them as deliberate attempts to suppress China's technological development.
The episode escalates the broader US-China competition for dominance in artificial intelligence — a domain both countries consider strategically vital.

Static Topic Bridges

DeepSeek — What It Is and Why It Matters

DeepSeek is a Chinese artificial intelligence research company that released a series of highly capable open-source language models, most notably DeepSeek-R1 in January 2025, which demonstrated reasoning capabilities matching those of OpenAI's frontier model o1 at a fraction of the training cost.

DeepSeek-R1 matched or came close to matching OpenAI's o1 on key benchmarks: GPQA (graduate-level science), AIME (advanced mathematics), and Codeforces (competitive programming)
Training cost reportedly 27 times lower than comparable US models — a major demonstration of efficiency
Smaller "distilled" versions of DeepSeek-R1 can run on consumer-grade laptops, bypassing the need for large data centres
Open-source release made it globally accessible — a stark contrast to closed US frontier models
When early versions were asked "what model are you?", they reportedly responded "I'm ChatGPT" — cited by OpenAI as direct evidence of distillation from its proprietary systems
"Distillation" in AI refers to training a smaller model using outputs generated by a larger proprietary model — a technique that can transfer capabilities without direct access to model weights

Connection to this news: DeepSeek's capabilities at low cost fundamentally challenged the assumption that US export controls on advanced chips would limit Chinese AI development — making it both a technological and geopolitical event of the first order.

US Export Controls on AI Chips — Background

Beginning in October 2022, the United States imposed a series of export controls designed to prevent China from acquiring advanced semiconductor chips necessary for training frontier AI models.

October 2022 controls: Restricted exports of advanced GPUs (including NVIDIA A100 and H100 chips) to China; required NVIDIA to develop restricted variants (A800, H800) for the Chinese market
These restrictions were tightened in October 2023 and again in October 2024, closing loopholes in performance thresholds
DeepSeek reportedly trained its models partly on stockpiled chips acquired before controls tightened, and partly on less advanced chips through highly efficient training algorithms
The Brookings Institution and RAND Corporation noted that DeepSeek demonstrated the limits of chip-based export controls — scarcity can incentivise innovation
Wassenaar Arrangement (1996): A multilateral export control regime covering dual-use goods and technologies; 42 member states (India is a member since 2017) coordinate on technology transfer restrictions. However, China is not a member, limiting the arrangement's effectiveness as a containment tool

Connection to this news: DeepSeek's emergence despite US chip controls directly challenged the strategic assumption underlying US AI policy — that controlling hardware access would maintain a decisive lead in AI capabilities.

Intellectual Property Rights in International Law — AI Context

The allegation that DeepSeek used "knowledge distillation" from OpenAI's models raises unresolved questions about intellectual property protection in the age of large language models.

TRIPS Agreement (Trade-Related Aspects of Intellectual Property Rights, 1995): WTO agreement establishing minimum standards for IP protection globally; covers copyright, patents, trade secrets — but its application to AI-generated outputs and AI training data is contested
WIPO (World Intellectual Property Organization): Has been developing frameworks for AI and IP since 2019; the WIPO Conversation on AI and IP has not yet produced binding standards on distillation or training data
In the US context, "knowledge distillation" — training a model on outputs of another model — occupies a legal grey area; courts are still determining whether AI outputs are copyrightable
OpenAI's Terms of Service prohibit using ChatGPT outputs to train competing models, but the enforceability of such terms across international jurisdictions is legally uncertain
China's domestic AI regulations (2023) require labelling of AI-generated content and data security compliance, but do not address outbound IP concerns

Connection to this news: The State Department's diplomatic campaign reflects a US attempt to build international normative pressure around AI IP protection in the absence of binding multilateral frameworks — an exercise in norm-setting rather than legal enforcement.

US-China Technology Rivalry — Strategic Dimensions

Artificial intelligence has become the central domain of the US-China strategic competition, with both governments treating AI leadership as a matter of national security.

US: CHIPS and Science Act (2022) — $52.7 billion for domestic semiconductor manufacturing; National AI Initiative Act (2020); Executive Order on AI Safety (October 2023); AI Safety Institute under NIST
China: "New Generation AI Development Plan" (2017) — targets global AI leadership by 2030; domestic champions include Baidu, Alibaba, Tencent, Huawei, and now DeepSeek; significant state investment in AI R&D
Export Control Reform Act (ECRA, 2018): US law that updated export control infrastructure to address "emerging and foundational technologies" — foundational for AI-related chip restrictions
The China Select Committee (House of Representatives) has been the primary US legislative body addressing Chinese technology threats, receiving OpenAI's February 2026 memo

Connection to this news: The State Department cable is part of a broader US diplomatic campaign to build allied consensus around technology containment — extending the bilateral US-China AI competition into a multilateral diplomatic contest.

India's AI Policy and Strategic Positioning

India occupies a distinctive position in the US-China AI competition — as a large democracy with a growing AI ecosystem, and a member of the Wassenaar Arrangement, India faces choices about technology alignment and AI governance.

NITI Aayog National Strategy for AI (2018): First comprehensive policy framework — focused on healthcare, agriculture, education, smart cities, and smart mobility; coined the phrase "AI for All"
India AI Mission (March 2024): Cabinet approval with a budget outlay of ₹10,371.92 crore over five years; focus areas include computing infrastructure, datasets, innovation centres, and responsible AI
Wassenaar Arrangement membership (since 2017): India is aligned with US-led export control regimes on dual-use technologies, though it maintains strategic autonomy in technology partnerships
India uses AI products and cloud services from both US companies (Google, Microsoft, AWS) and has not banned Chinese AI applications outright — in contrast to the US
India's AI regulatory approach remains comparatively lighter-touch, with a focus on innovation rather than restriction

Connection to this news: The US-China AI IP dispute is directly relevant to India's technology governance choices — India must navigate between US pressure on technology alignment and its own interest in accessing the most cost-effective AI tools for development, including potentially open-source Chinese models.

Key Facts & Data

State Department cable date: 24 April 2026
Companies named: DeepSeek, Moonshot AI, MiniMax (all Chinese)
OpenAI memo: Sent to US House China Select Committee, February 2026
DeepSeek-R1 release: January 2025
Training cost advantage: DeepSeek-R1 reportedly 27 times cheaper than OpenAI o1
US chip export controls: October 2022 (initial); tightened October 2023, October 2024
Chips restricted: NVIDIA A100, H100 class GPUs
Export Control Reform Act (ECRA): 2018 — legal basis for emerging technology controls
Wassenaar Arrangement: Established 1996; 42 members; India joined 2017; China not a member
TRIPS Agreement: 1995, WTO; covers IP standards but application to AI training is unsettled
CHIPS and Science Act (US): 2022; $52.7 billion for domestic semiconductor production
India AI Mission: Approved March 2024; budget ₹10,371.92 crore over 5 years
China's AI leadership target: 2030, per New Generation AI Development Plan (2017)