White House accuses China of ‘industrial scale’ AI theft

Source: ft.com

TL;DR

The story at a glance

The Financial Times reports the White House accusing China of "industrial scale" theft of AI technology amid escalating US concerns over Chinese efforts to close the AI gap. Key players include US AI firms like Anthropic and OpenAI, which have flagged Chinese labs such as DeepSeek, Moonshot AI, and MiniMax for distillation attacks, and White House officials like David Sacks highlighting IP risks. This is being reported now due to recent public allegations from AI companies and congressional hearings on Chinese AI espionage in early 2026.[[5]](https://www.latimes.com/business/story/2026-02-13/openai-accuses-chinas-deepseek-of-stealing-ai-technology)[[1]](https://www.ft.com/content/abde4e1e-c69a-4cc4-ad96-d88308314298)

Distillation involves training smaller models on outputs from larger ones like Claude, allegedly done illicitly at massive scale to bypass development costs.

Key points

Details and context

The article focuses on US accusations against China for large-scale AI IP theft via "distillation," where Chinese firms query US models millions of times to replicate capabilities cheaply, violating terms of service and export controls.[[3]](https://nathanbenaich.substack.com/p/state-of-ai-april-2026-newsletter)

This fits a pattern of economic espionage, including chip smuggling ($2.5bn Nvidia GPUs) and insider thefts, as China lags ~7 months behind US frontier AI but uses illicit means to catch up.[[4]](https://docs.house.gov/meetings/ZS/ZS00/20260416/119165/HHRG-119-ZS00-Wstate-MahmoodY-20260416.pdf)

Distillation circumvents US bans on advanced chips, allowing China to build competitive models like DeepSeek-R1 without full R&D costs, raising fears of military applications.[[5]](https://www.latimes.com/business/story/2026-02-13/openai-accuses-chinas-deepseek-of-stealing-ai-technology)

US responses include proposed AI-ISAC for threat sharing and congressional pushback, but no AI-ISAC timeline yet; China denies, calling accusations smears.[[9]](https://www.fdd.org/analysis/2026/04/08/the-american-ai-sector-bands-together-to-stop-chinese-theft)

Key quotes

"DeepSeek used OpenAI’s model to train its competitor using ‘distillation,’ White House AI czar says." — Beatrice Nolan, Fortune, cited in congressional testimony.[[4]](https://docs.house.gov/meetings/ZS/ZS00/20260416/119165/HHRG-119-ZS00-Wstate-MahmoodY-20260416.pdf)

"Steal, copy and erase — that is the Chinese Communist Party’s standard playbook." — John Moolenaar, House Select Committee chair.[[10]](https://economy.ac/news/2026/04/202604288808)

Why it matters

China's alleged theft undermines US AI dominance, potentially handing Beijing advantages in economic, military, and intelligence domains where AI is pivotal.

For US AI firms, it means billions in lost R&D value; investors and policymakers face pressure to protect IP without stifling innovation, while consumers may see cheaper Chinese models flood markets.

Watch for US sanctions on named firms, enforcement of export controls, and outcomes from House bills, though effectiveness depends on global cooperation amid denials from China.[[11]](https://www.businesstimes.com.sg/companies-markets/telcos-media-tech/us-house-weigh-penalties-ai-model-copying-chinese-firms)

What changed

(1) Previously, US AI firms developed models independently amid export bans on chips to China. (2) Now, Chinese labs allegedly bypass via distillation, stealing outputs at scale. (3) Allegations surfaced publicly in January-February 2026, with Anthropic report on February 23.[[3]](https://nathanbenaich.substack.com/p/state-of-ai-april-2026-newsletter)

FAQ

Q: What is AI distillation in this context?

A: Distillation trains a smaller AI model on outputs from a larger one like Claude to replicate capabilities. Anthropic claims Chinese firms did this illicitly with millions of queries via fake accounts, violating terms and evading regional blocks. It lets rivals shortcut costly training.[[2]](https://www.wsj.com/tech/ai/anthropic-accuses-chinese-companies-of-siphoning-data-from-claude-63a13afc)

Q: Which Chinese companies were accused?

A: Anthropic named DeepSeek, Moonshot AI, and MiniMax. OpenAI focused on DeepSeek using obfuscated methods to distill from its models. These firms created ~24,000 fake accounts for the attacks.[[3]](https://nathanbenaich.substack.com/p/state-of-ai-april-2026-newsletter)

Q: How has the White House responded?

A: AI advisor David Sacks confirmed evidence of DeepSeek distillation from OpenAI, warning it slows copycats but signals IP theft risks. The administration pushes AI Action Plan for threat monitoring, but specifics on sanctions remain pending.[[6]](https://ca.finance.yahoo.com/news/white-house-evaluates-china-ai-185638531.html)

Q: What US actions are proposed?

A: House bill "Deterring American AI Model Theft Act" would blacklist firms using query-copy tactics, impose sanctions, and create threat intel hubs. Hearings urge faster AI-ISAC setup.[[12]](https://www.instagram.com/p/DXM0kQWDZqy)