White House Accuses Chinese Firms of AI Theft as US Tech Giants Unite to Combat Espionage

The White House has accused Chinese firms of conducting industrial-scale theft of American AI models through a technique known as distillation, prompting commitments to share intelligence with leading U.S. companies like OpenAI, Anthropic, and Google. A policy memo from the White House Office of Science and Technology Policy (OSTP), authored by Michael Kratsios, highlights this as a systematic effort to replicate U.S. AI capabilities without permission, as reported by BBC Business and The Next Web.

U.S. AI developers have been vocal about specific incidents fueling these concerns. OpenAI publicly accused DeepSeek of distilling its models back in February, while Anthropic detailed how DeepSeek, MiniMax, and Moonshot AI created over 24,000 fraudulent accounts to generate 16 million interactions with its Claude models, violating terms of service. Google has similarly reported distillation campaigns targeting its Gemini model, including one with over 100,000 prompts aimed at copying reasoning abilities in non-English languages, according to analyses from the Foundation for Defense of Democracies and Computer Weekly.

China has dismissed the allegations as slander, according to Ars Technica, amid rising tensions that could overshadow an upcoming Trump-Xi summit. The accusations extend beyond software to hardware, with U.S. officials uncovering massive chip-smuggling operations—such as a $2.5 billion scheme involving altered serial numbers and fake servers—to evade export controls and fuel China's AI ambitions, as outlined in a recent Select Committee on China hearing.

In response, American AI giants—typically fierce competitors—are banding together. OpenAI, Anthropic, and Alphabet (Google's parent) announced plans through their Frontier Model Forum to collaborate on detecting and publicly reporting Chinese-linked espionage, marking a rare unity in the sector. The White House memo pledges government intelligence sharing with these firms and exploration of accountability measures, signaling a push for stronger defenses.

This dispute matters deeply because AI underpins economic and military power. Distillation allows "student" models to mimic advanced "teacher" ones at low cost, eroding U.S. leads without needing huge training investments or direct hacks. Chinese models, often cheaper or free, pose business risks to American firms while raising national security fears over weaponized technology.

Those affected include U.S. AI developers losing proprietary edges, potentially slowing innovation, and global users relying on secure AI tools. China, still dependent on U.S. chips and tech across the AI stack, faces sanctions risks that could disrupt its rapid catch-up.

Looking ahead, Congress is urged to bolster protections, as emphasized in hearings, while private-sector cooperation grows but may need more government oversight. The Trump-Xi talks loom as a potential flashpoint, with the U.S. mulling major penalties to safeguard its technological lead.