Stealing AI's Answer Key: How the US-China Tech War Just Escalated
openrouter:bytedance-seed/seed-2.0-lite · prompt: edited
· 2026-05-30T11:45:05 · 1.73¢
· 🔍 view original input ↗
📃 Rewritten passage
In April 2026, the White House launched one of the sharpest escalations yet in the US-China race to dominate artificial intelligence, issuing a formal memo accusing Chinese entities of running industrial-scale, unauthorized campaigns to steal American frontier AI technology. The memo, penned by White House Office of Science and Technology Policy director Michael Kratsios, frames the practice as a deliberate attack on US innovation, released just weeks before a scheduled summit between US President Donald Trump and Chinese leader Xi Jinping in Beijing. At the core of the accusations is model distillation, a process that can be used legally or illegally. Legitimate distillation shrinks large, expensive AI models into smaller, more accessible versions for everyday use, a practice Kratsios acknowledges is a vital part of the global AI ecosystem. Unauthorized distillation, by contrast, scrapes outputs from leading US models to train competing Chinese models at a fraction of the cost of building a frontier model from scratch. Kratsios’ memo details how Chinese actors use tens of thousands of proxy accounts and AI jailbreaking techniques to evade detection while accessing proprietary US model outputs. The accusations are not unprecedented: in early 2025, OpenAI accused leading Chinese AI firm DeepSeek of using outputs from its GPT models to train DeepSeek’s own platform, violating OpenAI’s terms of service. In February 2026, Anthropic extended those accusations, naming DeepSeek alongside two other top Chinese AI firms—Moonshot and MiniMax—for running similar distillation attacks on its models. Chinese officials have pushed back hard on the claims, with embassy spokesperson Liu Pengyu calling the White House’s accusations “pure slander.” Liu noted that China has committed to intellectual property protections and healthy, cooperative tech development. But US officials and AI firms argue the practice erodes the competitive edge the US has built via years of research and strict export controls on advanced AI chips. Council on Foreign Relations tech security expert Chris McGuire notes that Chinese firms use distillation to offset gaps in access to the high-powered computing hardware needed to train frontier models, allowing them to close the AI gap without making the same massive investments US firms have. US AI leaders add another critical risk: distilled models lack the built-in safety safeguards that prevent original US models from being used to develop bioweapons or launch malicious cyberattacks. In response to the ongoing threats, the House Foreign Affairs Committee has passed a series of bills that would add firms caught using unauthorized distillation to the US Entity List, a trade blacklist that blocks American companies from selling technology to listed entities. The White House memo also pledges to share intelligence on distillation attempts with US AI firms and coordinate new measures to hold foreign actors accountable. While Kratsios admits that unauthorized distilled models do not match the performance of the original US models, their low cost and ability to erode US competitive advantage make them a core flashpoint in the accelerating global AI race.
The US-China AI arms race just took a new turn, with Washington accusing Beijing of mass, underhanded theft of American frontier tech.
📖 What's Going On?
The White House issued a formal April 2026 memo accusing Chinese entities of running industrial-scale, unauthorized distillation campaigns to copy American frontier AI models. OSTP director Michael Kratsios’ memo alleges Chinese actors used tens of thousands of proxy accounts and jailbreaking to avoid detection while scraping US model outputs.
The accusations come weeks before a Trump-Xi summit, marking the latest escalation in the global AI dominance race. US leaders OpenAI and Anthropic previously flagged this practice, with Anthropic naming three Chinese firms for distillation attacks on its models in February 2026.
🎯 How To Think About It
Unauthorized AI distillation is a reverse-engineered cheat code, and two analogies make its unfair, cost-cutting advantage clear:
- It’s like sneaking into an MIT research lab, copying every graduate-level final exam answer key, and using those keys to train your own entry-level students to outperform peers who learned material from scratch, at 1% of the cost of building your own lab.
- It’s equivalent to a competitive gaming team scraping every pro player’s in-game strategy from Twitch to train an amateur team’s AI bot, letting them beat high-level opponents without spending years building their own training infrastructure.
💡 Key Things To Know
- Legitimate distillation is a standard, legal AI practice used to shrink large models for accessible consumer use, per Kratsios’ memo.
- Unauthorized distillation lets Chinese firms avoid the hundreds of millions of dollars it costs to train frontier models from scratch, closing the US’s competitive edge.
- The US Entity List, a trade blacklist, is a core penalty the government could use against firms caught running unauthorized distillation campaigns.
- Most people assume distilled models match the original’s power, but the memo notes they lag in performance while still eroding US leadership.
- American firms warn distilled models lack built-in safety safeguards that prevent use for bioweapons development or malicious cyberattacks.
🌟 Why It Matters
This AI arms race is not a distant geopolitical drama—it will shape the career paths you choose, the education tools you use, and the safety of the digital spaces you navigate daily. AI is already a $1 trillion global industry, and US leadership in the space will determine what entry-level roles open up for computer science, engineering, and policy students over the next decade. Unregulated, unsafe AI could also reshape social media, education, and national security in ways that directly impact your ability to use tech safely as you enter adulthood.
🔮 The Bigger Picture
The latest accusations are set to dominate the upcoming Trump-Xi summit, and their outcome will set global AI regulation rules for the next decade. Watch for new export controls that limit foreign access to US models, a surge in domestic Chinese AI investment to avoid reliance on US tech, and a global split into two incompatible AI ecosystems for consumers and businesses.
📖 Glossary (5)
- Frontier AI — The most advanced, capable large language models (LLMs) built by leading tech firms, which require billions of dollars in computing power to train.
- Model distillation — A process that trains a smaller, cheaper AI model on the outputs of a larger, more powerful "teacher" model to replicate its core capabilities.
- AI jailbreaking — A technique that bypasses an AI model’s built-in safety and access controls to access proprietary information or generate restricted content.
- Entity List — A US government trade blacklist that prohibits American companies from selling technology to listed foreign entities without special government approval.
- Proxy accounts — Fake or third-party user accounts used to hide the true identity of someone accessing an online service, most often to avoid detection for rule-breaking.
📝 Quiz (10) — with answers
Q. The passage’s central claim is most accurately stated by which choice?
A. Chinese AI firms have successfully overtaken US AI leadership through unauthorized model distillation
B. The US government’s accusations of Chinese AI IP theft mark a major escalation in the global AI arms race ✓
C. The Trump administration will impose immediate sanctions on all Chinese firms accused of distillation
D. Model distillation is an illegal practice that has no legitimate uses in the global AI ecosystem
Answer: B — The passage centers the White House’s 2026 memo as the latest escalation in US-China AI competition, making B correct. The most tempting wrong answer is D, a Trap A that reverses the passage’s clarification that legal distillation is a common industry practice. SAT Tip: For central idea questions, eliminate options that contradict explicit details in the passage to narrow down viable choices quickly.
Q. According to the passage, Kratsios states that unauthorized distillation differs from legitimate distillation in that it
A. produces models that outperform original US frontier AI models
B. requires tens of thousands of dollars in computing infrastructure
C. undermines US AI research and development to gain a competitive edge ✓
D. is used exclusively by Chinese firms to shrink large models for consumer use
Answer: C — Kratsios notes industrial-scale distillation used to erode American R&D is unacceptable, while legitimate distillation is a valid part of the AI ecosystem, making C correct. The main distractor is A, a Trap C that contradicts the passage’s claim that distilled models underperform originals. SAT Tip: For detail-focused questions, cross-reference every option with explicit text in the passage to avoid misremembering author claims.
Q. The passage notes that Anthropic accused which of the following groups of distillation attacks on its models?
A. Three leading Chinese AI companies ✓
B. White House science policy officials
C. American chip manufacturers exporting to China
D. OpenAI and other US AI firms competing with Anthropic
Answer: A — The passage states Anthropic named DeepSeek, Moonshot, and MiniMax—three top Chinese AI firms—for distillation attacks, so A is correct. The main wrong option is D, a Trap B that misrepresents US AI firms as allies, not targets, of accusations. SAT Tip: If a question asks for a specific named group, scan the passage for proper nouns to locate the exact detail before choosing an answer.
Q. As used in the passage, the word “principal” (to describe Chinese entities) most nearly means
A. leading ✓
B. educational
C. minor
D. financial
Answer: A — The passage uses "principal" to describe the main, leading group of foreign entities accused of distillation, so "leading" is the correct in-context meaning. The most tempting wrong answer is B, the common colloquial use of "principal" for a school leader, a Trap C that misapplies a common definition. SAT Tip: For vocab-in-context questions, replace the original word with each option to test which one preserves the sentence’s original meaning.
Q. As used in the passage, the word “surreptitious” (to describe distillation campaigns) most nearly means
A. quick
B. secret ✓
C. widespread
D. expensive
Answer: B — "Surreptitious" describes unauthorized campaigns that hide their activity via proxy accounts, so "secret" matches the in-context meaning. The main distractor is C, a Trap B that misapplies the passage’s description of campaigns as "industrial-scale" (widespread) rather than their core hidden nature. SAT Tip: Always check surrounding sentences for context clues that explain the word’s use, rather than relying on your first impression of its definition.
Q. The passage suggests that US AI firms view unauthorized distillation as a threat primarily because it
A. eliminates their ability to train small, low-cost models for consumer use
B. erodes their competitive advantage and creates unregulated safety risks ✓
C. forces the US government to impose new export controls on American chips
D. leads to a global reduction in investment in AI research and development
Answer: B — The passage cites US firms’ concerns that distillation closes their competitive edge and that distilled models lack critical safety safeguards, making B correct. The main wrong option is C, a Trap A that reverses cause and effect: distillation leads to potential new export controls, not the other way around. SAT Tip: For cause-effect inference questions, map the order of events in the passage to avoid mixing up triggers and outcomes.
Q. Which statement about the upcoming Trump-Xi summit can most reasonably be inferred from the passage?
A. The summit will result in a permanent end to US-China AI tensions
B. The AI distillation accusations will be a key topic of discussion during the meeting ✓
C. China will issue a formal apology for its distillation campaigns at the summit
D. The summit will lead to new global standards for legitimate AI model distillation
Answer: B — The passage notes the accusations come just weeks before the summit, implying the issue will be a core agenda item, so B is correct. The main distractor is D, a Trap C that is a plausible real-world outcome but not supported by any explicit or implied claim in the passage. SAT Tip: For inference questions, only choose claims that are directly supported by passage text, not just logical-sounding real-world possibilities.
Q. The author’s primary purpose in writing the passage is to
A. advocate for harsh US sanctions on all Chinese AI firms
B. explain a new development in the US-China global AI competition ✓
C. dismiss the White House’s accusations as unfounded political rhetoric
D. compare the capabilities of American and Chinese frontier AI models
Answer: B — The passage’s core goal is to inform readers about the White House’s new accusations and their place in the broader US-China AI race, making B correct. The main wrong option is A, a Trap B that misattributes a persuasive stance to the neutral, explanatory author. SAT Tip: Most SAT nonfiction passages are explanatory, focused on presenting information rather than arguing a personal position—use that to eliminate persuasive purpose options first.
Q. The passage indicates that US officials view distilled AI models as which of the following?
A. Superior to the original US frontier models they replicate
B. A threat to US AI leadership but not capable of matching original models ✓
C. Eligible for inclusion in legitimate AI ecosystem practices
D. Too expensive to be a viable long-term competitor for US models
Answer: B — Kratsios’ memo states distilled models do not match original performance but still erode US competitive advantage, making B correct. The main distractor is A, a Trap A that reverses the passage’s explicit claim that distilled models underperform originals. SAT Tip: For questions about official stances, locate the section of the passage that quotes the relevant official’s comments to confirm your answer.
Q. Which choice provides the best evidence for the answer to the previous question?
A. "the White House has accused China of undertaking industrial-scale theft of American artificial intelligence labs’ intellectual property"
B. "Chinese campaigns were “leveraging tens of thousands of proxy accounts to evade detection and using jailbreaking techniques to expose proprietary information”"
C. "while models created by “surreptitious, unauthorised distillation campaigns” did not match the performance of the original models, they can benefit foreign groups because of the significantly lower cost" ✓
D. "American AI companies are concerned that distilled models pose national security risks because they lack the safeguards that, for example, prevent the development of bioweapons"
Answer: C — Option C directly quotes the passage line that states distilled models lag in original performance but provide cost benefits that threaten US leadership, matching the answer to Q9. The main distractor is D, a Trap B that cites a separate safety concern, not the relative performance of distilled models. SAT Tip: For evidence-pairing questions, answer the first question first, then locate the exact line in the passage that supports your answer before selecting the matching option.
💬 Suggested questions
- Why can't US AI firms block all proxy accounts used to distill their models?
- What would happen if the US added DeepSeek to the Entity List tomorrow?
- How does model distillation actually work to copy a large AI's core capabilities?
Raw JSON