AI cold war heats up: US slams China for 'industrial-scale theft' as beijing fires back with investment bans

By early 2026, the US State Department has intensified its allegations against Chinese AI companies such as DeepSeek, Moonshot AI, and MiniMax, accusing them of large-scale intellectual property theft through sophisticated techniques like model extraction and distillation. Unlike traditional IP theft involving direct source code copying, these firms reportedly employ proxy accounts to clandestinely distill advanced American AI models, including Anthropic's Claude, thereby circumventing existing restrictions and complicating enforcement efforts.

This model distillation tactic enables Chinese AI labs to effectively bootstrap their own systems by leveraging the strengths of superior US models, particularly enhancing weaker capabilities such as coding proficiency. While this approach accelerates China’s AI development trajectory, it still relies on foundational American models as a starting point, highlighting a nuanced form of intellectual property misuse that blurs the lines between innovation and infringement.

Despite mounting evidence of unauthorized distillation, US policymakers face significant challenges in quantifying its precise impact on China’s AI advancement, making it difficult to craft targeted enforcement strategies. As experts note, the speed and stealth of these operations raise concerns about an accelerating AI arms race, with the net effect of this covert knowledge transfer remaining elusive and complicating diplomatic and regulatory responses.

The United States has escalated its diplomatic posture by publicly accusing China-backed actors, including firms like DeepSeek, of conducting industrial-scale intellectual property theft through sophisticated model distillation campaigns that siphon capabilities from American AI labs. The White House’s framing of these actions as “industrial scale theft” signals a new confrontational phase, with officials preparing enforcement measures aimed at holding foreign actors accountable, reflecting a readiness to pursue legal and regulatory actions against Chinese AI firms. This hardening stance underscores the intensifying rivalry as the US seeks to protect its AI innovations from what it views as systematic siphoning.

In response, China has implemented countermeasures such as banning foreign investment in AI companies with Chinese founders abroad, exemplified by its prohibition of Meta’s acquisition of the Singapore-based AI firm Manis, founded by Chinese entrepreneurs. This regulatory move not only forces Meta to unwind its December 2025 deal but also signals Beijing’s intent to shield its AI sector from foreign influence amid rising diplomatic tensions. Meanwhile, China vehemently denies the US accusations, labeling them as “entirely baseless” and a “slanderous smear” against its AI industry, highlighting the deepening diplomatic acrimony and the complex interplay between enforcement and political posturing.

These diplomatic and enforcement developments are unfolding against the backdrop of a high-stakes US-China summit scheduled for next month, where AI competition and security concerns loom large. The summit underscores the geopolitical significance of AI rivalry, as both nations grapple with the dual imperatives of advancing AI capabilities while managing escalating accusations and enforcement posturing. This convergence of diplomacy and enforcement reflects a more aggressive and complicated phase of the AI competition, where technological leadership and national security are increasingly intertwined.

U.S. export controls on advanced semiconductor chips remain a pivotal strategic lever in the AI rivalry with China, effectively constraining Beijing's ability to develop large-scale, frontier AI capabilities such as Mythos-class models. Despite internal industry pressures to relax these controls due to lost revenue from the Chinese market, policymakers recognize that maintaining these restrictions is essential not only for national security but also for preserving America's economic and technological leadership during this critical 'breakaway opportunity' period around 2026. As underscored by analyses from May 2026, these export controls are a rapidly appreciating asset rather than a sunk cost, with the Biden Administration's containment strategy explicitly aiming to slow China's progress toward artificial general intelligence (AGI) by using chips as a chokepoint, thereby preserving a projected six-month U.S. lead in frontier AI capabilities by 2028.

The evolving U.S.-China AI competition transcends a simple race for technological supremacy, encompassing a multifaceted contest over global AI norms, values, and influence. While the U.S. focuses on commercial ventures like Anthropic and OpenAI with closed-source models, China aggressively advances open-source AI ecosystems, producing efficient, low-cost models such as Xiaomi's GLM that are rapidly closing the performance gap and gaining widespread adoption—even within U.S. startups and academia. This dynamic challenges U.S. policy frameworks that prioritize export controls and IP enforcement, as Chinese firms like DeepSeek employ large-scale model distillation techniques—described by Chinese state media as a 'back door'—to replicate U.S. innovations, blurring lines between legitimate competition and intellectual property theft. Consequently, the U.S. faces the dilemma of balancing national security concerns with fostering innovation and managing dependencies on Chinese open-source contributions.

Managing the AI rivalry requires not only technological containment but also sophisticated geopolitical and diplomatic strategies akin to Cold War crisis planning. Thought leaders like Anthropic’s Jack Hirsch emphasize the necessity of establishing communication channels to mitigate existential risks and geopolitical tensions, highlighting that U.S. policy must plan for both competition success and post-leadership global stability. The Biden Administration’s approach reflects this dual imperative by promoting an 'entente strategy' among allied liberal democracies to scale AI capabilities rapidly while restricting authoritarian rivals’ access to critical technologies. However, this containment approach complicates U.S. policy by potentially stifling open-source innovation domestically and creating control gaps, underscoring the need to target regulation at bad actors rather than broadly restricting AI development.

The rapid evolution of AI model frontiers demands a dynamic and adaptive U.S. AI ecosystem that can evolve independently of reliance on Chinese technology stacks, which may be optimized for their own hardware and embed foreign dependencies. As Google CEO Sundar Pichai notes, the fast-changing nature of AI models necessitates building flexible infrastructures capable of adapting to shifting technological landscapes. This challenge is compounded by the U.S. AI ecosystem’s trend toward closed-source models controlled by a few companies, potentially slowing innovation and ceding ground to China’s open-source leadership. Addressing these challenges requires nuanced policy that balances safeguarding technological leadership with fostering innovation and mitigating risks of dependency on adversarial AI ecosystems.