Distillation Attacks: The Frontier of AI Warfare

AI Distillation Attacks: Inside China's Race to Extract U.S. AI Capabilities

A new EdgeTheory assessment on the fastest-growing flashpoint in the U.S.-China AI race

AI distillation - the technique of training a smaller "student" model on a powerful "teacher" model's outputs - has become one of the most contested issues in global AI competition. Anthropic and OpenAI allege that Chinese firms including Moonshot AI, DeepSeek, MiniMax, and Alibaba have run industrial-scale campaigns to extract reasoning, coding, and agentic capabilities from U.S. frontier models like Claude - using tens of thousands of fraudulent accounts and tens of millions of exchanges. At the same time, PLA-linked research institutions have published work applying distillation to military and security applications.

EdgeTheory's latest assessment cuts through the noise to separate legitimate model distillation from suspected unauthorized extraction, and explains why this fight matters for the future of U.S. AI leadership.

What You'll Learn

• How distillation attacks actually work - the technical mechanics behind how a "student" model learns from a "teacher" model's outputs, and why reasoning traces and agentic trajectories are far more valuable to extract than simple answers.
• Who's behind the alleged campaigns - a breakdown of the key suspected actors, including Moonshot AI, DeepSeek, MiniMax, Alibaba/Qwen, and PLA-linked research institutions, and what's actually been proven versus alleged.
• Why this threatens the U.S. AI advantage - how distillation shifts the competition from controlling advanced chips to protecting model capabilities, and what that means for the future of export controls.
• The geopolitical fallout for the global AI race - how distillation could accelerate China's AI catch-up, expand its global influence, and drive deeper U.S.-China technology fragmentation.
• The real national-security stakes - how commercially developed AI capabilities could diffuse into military, cyber, surveillance, and intelligence applications, and what policymakers are already proposing in response.

Understanding these dynamics is no longer optional for anyone tracking U.S.-China technology competition.

As the fight over AI shifts from silicon to software, the actors, methods, and policy responses covered in this report will shape how the next phase of the AI race is won - or lost.

Download the report

The 2026 Cyclospora outbreak began with a relatively straightforward public-health question: What was making people sick?

But as the outbreak spread, the questions became much bigger.

Who was responsible? Was Taco Bell to blame? What did Taylor Farms' role say about the food supply chain? Was imported produce putting consumers at greater risk? And had government oversight failed to keep Americans safe?

The FDA provides the factual baseline. But the narratives surrounding those facts extend much further.

A specific lettuce outbreak becomes a question about the safety of fresh produce. Traceback to a Mexican supplier becomes concern about imported food. Exposure at Taco Bell becomes a debate over brand responsibility. And a complex federal investigation becomes evidence, for some audiences, that government oversight is failing.

Using EdgeTheory's C2 Narrative Intelligence Platform, this report examines the six major narratives shaping the 2026 Cyclospora outbreak—along with the sources and actors amplifying them, the emotions giving them momentum, and their potential implications for brands, consumer behavior and business performance.

Facts establish what happened. Narratives shape what those facts come to mean.

• How a food-safety outbreak became a much broader crisis of trust.
• Why Taco Bell, Taylor Farms, imported produce, and government oversight became part of the same narrative environment.
• Who helped shape and amplify the narratives—and which emotions gave them momentum.
• How narratives can move beyond online conversation to influence brand reputation, consumer behavior and business outcomes.

From Foodborne Outbreak to Crisis of Trust

The immediate health threat will eventually subside. The narratives it created may not.

Taco Bell can remain associated with illness. Taylor Farms can become associated with repeated food-safety incidents. Mexican produce can become part of broader arguments about imports. And regulatory difficulties can reinforce existing distrust of government institutions.

That distinction is central to Narrative Intelligence. Understanding not only what happened, but how people interpret what happened, who shapes those interpretations, and how they spread, can provide organizations with a clearer view of emerging brand, reputation and business risk.

The most important question, then, is no longer only Is this lettuce safe? It becomes:

Can I trust the nation's food supply?

Download the report

EdgeTheory Lab and the National Center for Narrative Intelligence (NCNI) are now offering the Certified Professional in Narrative Intelligence (CPNI) - the only credential of its kind, established through NCNI's congressional mandate under the NDAA.

Learn more about the CPNI program.

Most supply chain risk tools monitor physical disruptions after they're reported. Narrative intelligence detects the signals forming days earlier. Here's what that gap actually looks like.

New Narrative Intelligence Report

This EdgeTheory report examines the distinction between AI Sovereignty and Sovereign AI, using Narrative Intelligence (NARINT) analysis to show how governments, industry, and media frame these competing concepts to advance different strategic goals around technological autonomy. The report finds that AI Sovereignty (control and governance over AI systems and data) is primarily promoted by Western actors, while Sovereign AI (domestic infrastructure, compute, and models) is primarily advanced by China, Russia, India, and other emerging powers - fragmenting the global AI market into regionally aligned ecosystems.

What you'll learn

• The core distinction — AI Sovereignty is the goal of controlling AI governance and data, while Sovereign AI is the means of achieving that goal through domestically built infrastructure, compute, and models.
• Why the Houthi Movement resumed attacks on Saudi Arabia and imposed a maritime embargo, and how this serves Iranian strategic interests
• Who's promoting which narrative — AI Sovereignty coverage is dominated by Chinese-aligned (43.8%) and Russian-aligned (25%) sources, while Sovereign AI coverage is similarly China-heavy (46.7%) but with stronger Indian representation (26.7%) and little Russian presence.
• Emotional framing differs by narrative — Sovereign AI content is almost exclusively fear-driven (97%), signaling urgency around building domestic capability, while AI Sovereignty content mixes fear (58%) and anger (39%), reflecting frustration over foreign dependence and governance failures.
• Country-specific positioning — China is best-positioned for Sovereign AI via open-weight models and Global South partnerships; Europe leads on AI Sovereignty through regulation (e.g., the EU AI Act) but remains dependent on U.S. chips and cloud; India is pursuing a hybrid strategy; Russia faces semiconductor and isolation constraints despite strong political will.
• Geopolitical and market stakes — Both narratives aim to reduce dependence on U.S. technology, which is fragmenting the global AI market into regionally distinct ecosystems and could gradually erode U.S. leverage over infrastructure, standards, and supply chains — even as narratives largely remain low-incitement and evidence-based, with isolated exceptions like Geopolitics & Empire's high-incitement framing.

Download the report

New Narrative Intelligence Report

This EdgeTheory report examines how the collapse of the US-Iran ceasefire framework in July has reshaped the regional conflict, with a particular focus on Iraq's emerging role as a critical battleground. As fighting between the US, Iran, and their respective allies intensified—drawing in Saudi Arabia, the Houthis, and Iran-backed Iraqi militias—Baghdad has become central to US efforts to curb Iranian influence through disarmament negotiations, anti-corruption crackdowns, and billions in new economic investment.

What you'll learn

• How the US-Iran memorandum of understanding broke down beginning July 7, and the strikes, blockades, and retaliatory attacks that followed
• Why the Houthi Movement resumed attacks on Saudi Arabia and imposed a maritime embargo, and how this serves Iranian strategic interests
• How Iran-backed Iraqi militias have continued attacks—including a bounty offer targeting President Trump—despite a broader informal pause in fighting
• What the "June Agreement" entails, and how Prime Minister Al-Zaidi's government has tied US investment to militia disarmament and financial crackdowns
• The scope and political fallout of Iraq's "Dawn Sweep" anti-corruption campaign, including which factions and officials were targeted
• Why Iraq's militias are unlikely to fully disarm by the September 30 deadline, and the risks this poses for Al-Zaidi's government
  How Saudi Arabia's direct entry into the conflict alongside the US signals a broader regional realignment against Iran

Download the report

This EdgeTheory report examines whether AI market bubble narratives are supported by evidence, finding that while valuation, debt, and commercialization risks are real, many bearish claims are amplified by Chinese state media and geopolitical actors seeking to undermine confidence in U.S. AI markets.

New Narrative Intelligence Report

Meta has become one of the world's most influential technology companies—and one of its most scrutinized. As the parent company of Facebook, Instagram, WhatsApp, and Threads, its decisions increasingly shape conversations about privacy, artificial intelligence, misinformation, corporate power, and public trust.

But reputations are rarely defined by a single event.

Using EdgeTheory's C2 Narrative Intelligence Platform, this report analyzes thousands of news articles and online conversations to uncover the five narrative ecosystems shaping Meta's public reputation—and introduces Narrative Convergence, EdgeTheory's Narrative Intelligence framework for understanding how independent conversations reinforce one another and evolve into lasting institutional beliefs.

What You'll Learn

• Why Meta's reputation extends beyond individual controversies.
• How five narrative ecosystems converge to shape institutional trust.
• Why fear and anger consistently reinforce reputational risk.
• How Artificial Intelligence amplifies existing trust narratives.
• Why Narrative Intelligence provides earlier warning than traditional media monitoring.
• How organizations can identify emerging narrative risks before they become strategic challenges.

Download the Report

Foreign intelligence services no longer need to recruit a single source to influence a decision — they can run coordinated narrative campaigns that reach an entire city council, legislature, or community at once. This piece looks at what a recent House Intelligence Committee hearing revealed about that shift, from a fabricated post that nearly derailed Florida counterintelligence legislation to organized opposition against AI data center projects, and explains why foreign influence operations are fundamentally a counterintelligence problem — plus how tools like EdgeWatch help decision-makers spot these narratives before they shape outcomes.

This EdgeTheory report examines the 14-point Memorandum of Understanding the US and Iran agreed to on 17 June, broadly favorable to Iran, and the future of Iraq's Iran-backed militias taking shape amid US scrutiny of government formation. It covers Iran's emerging "hybrid strategy" of integrating some militias into state security institutions while hardline factions like Kata'ib Hezbollah refuse to disarm, US Presidential Envoy Tom Barrack's conditions for supporting the Al-Zaidi government, and growing pressure to dismantle militias' financial networks ahead of Al-Zaidi's July meeting with President Trump.

The report analyzes competing narratives across state media, policy institutions, industry sources, independent outlets, social media, and activist networks, covering AI competition, environmental impacts, surveillance, and digital sovereignty.

Using EdgeRunner, an AI powered source discovery agent, the assessment combines Narrative Intelligence (NARINT), Emotional Profile Classification, and network analysis to reveal how governments, corporations, activists, and alternative media shape public perceptions of AI infrastructure.