China’s AI Heist: SHOCKING Tech Theft Unveiled

In,Data,Center,Two,Military,Men,Work,With,Open,Server

STOLEN AI Secrets: U.S. Security at Risk

Chinese AI firms are brazenly stealing American technology through mass distillation, handing the CCP a dangerous edge in the AI arms race that threatens U.S. national security.

Story Highlights

OpenAI accuses DeepSeek of using fake accounts and routers to extract ChatGPT outputs for its free R1 model, evading U.S. defenses.
Anthropic reveals 24,000 fake accounts siphoned 16 million Claude interactions; Google flags 100,000+ Gemini prompts stolen.
House Select Committee on China warns of CCP’s “steal, copy, kill” playbook undermining billions in U.S. AI investments.
Distilled models weaken safeguards, enabling censorship, cyber ops, and risks in biology—free-riding on American innovation.
President Trump’s administration eyes tougher controls to protect IP and counter China’s tech theft tactics.

OpenAI Exposes DeepSeek’s Theft Tactics

On February 12, 2026, OpenAI sent a memo to the House Select Committee on China detailing how DeepSeek extracted outputs from ChatGPT to train its R1 chatbot. DeepSeek employed third-party routers, fake accounts, and programmatic access to bypass defenses. This distillation process—training smaller models on larger ones’ responses—allowed China to replicate U.S. capabilities without massive infrastructure costs. Rep. John Moolenaar called it the CCP’s playbook: steal, copy, kill. American firms invested billions in frontier models while Chinese rivals offer free alternatives, eroding U.S. dominance.

OpenAI has warned US lawmakers that its Chinese rival DeepSeek is using unfair and increasingly sophisticated methods to extract results from leading US AI models to train the next generation of its breakthrough R1 chatbot https://t.co/siq46jRyM1

— Bloomberg (@business) February 12, 2026

Anthropic and Google Confirm Large-Scale Attacks

Anthropic’s February 23 report accused DeepSeek, Moonshot AI, and MiniMax of using 24,000 fake accounts for over 16 million Claude exchanges, targeting reasoning, coding, and tool use. Moonshot AI conducted 3.4 million queries; MiniMax hit 13 million. Google separately warned of over 100,000 prompts against Gemini. Anthropic’s Jacob Klein stated high confidence in scaled distillation yielding meaningful gains. These attacks persist despite U.S. firms’ blocks, with no silver bullet available. White House AI advisor David Sacks had flagged DeepSeek’s tactics in 2025.

National Security Risks from Weakened Models

Distilled Chinese models strip critical safeguards built into U.S. systems, posing risks for censorship, cyberattacks, disinformation, and sensitive biology applications. DeepSeek’s R1 censors topics like Taiwan and Tiananmen Square, contrasting U.S. open principles. This theft bypasses 2023 chip export controls on Nvidia H800s, as DeepSeek used 2.8 million GPU hours via Singapore proxies. Precedents include ex-Google engineer Linwei Ding’s 2026 conviction for stealing AI secrets. U.S. probes into chip flows intensify under President Trump’s America First agenda.

US-China AI Race Demands Action

U.S. firms like OpenAI and Anthropic charge for premium access to recoup investments, while Chinese labs free-ride to accelerate development. House Committee chair Moolenaar vows continued scrutiny of distillation threats. No responses emerged from Chinese firms or embassy. This escalation signals need for stronger IP protections and export policies. President Trump’s focus on countering CCP aggression aligns with shielding American innovation from theft that fuels adversarial advances in military and surveillance tech.

Protecting American Innovation

Conservatives celebrate pushback against China’s unfair tactics that undercut free enterprise and national security. Trump’s administration prioritizes confronting such overreach, echoing vows to restore U.S. tech supremacy. Limited Chinese rebuttals leave accusations unrefuted, with consensus on theft scale from multiple U.S. sources. Broader effects shift industry toward RL protections and question chip ban sufficiency, spurring government action to name offending labs and tighten controls.