The end of the 20th century was marked by the rise of the internet, one of history's greatest engines of globalisation. Territorial boundaries were flattened by open protocols like HTTP that allowed a developer in Bengaluru, a researcher in Silicon Valley and a student in Beijing to collaborate on a single digital canvas. This hyper-connected era birthed borderless software, global tech supply chains and a free exchange of human knowledge. It operated on one simple premise: code, ideas and data must flow freely to where they serve the highest utility. Recent policy shifts among leading AI powers hint at the world's gradual transition towards deglobalisation.
An algorithmic iron curtain
The most direct manifestation of this digital lockdown is the targeted restriction of access to frontier AI models. Cyberspace is shrinking as nations have begun restricting foreign access to their state-of-the-art systems to protect their competitive edge.
The United States has transitioned from a culture of open science to strict containment. The leading American labs, such as OpenAI and Anthropic, are implementing aggressive geo-blocking, cutting off cloud API (application programming interface) access to their advanced models for users in adversarial nations to prevent hostile exploitation. Beijing, meanwhile, has erected strict ideological barriers. Under its generative AI regulations, any model deployed domestically must strictly reflect the socialist core values. To maintain such an ideological framework, foreign players like OpenAI are blocked, while the domestic giants like Baidu’s Ernie Bot operate within a highly controlled national intranet.
On the other hand, the European Union AI Act has also created a defensive wall of compliance. By enforcing stringent copyright rules and data audits, Europe has effectively cordoned off its market. This has prompted Western tech firms to delay or entirely withhold the release of advanced multimodal tools in Europe, causing deeper fragmentation of the global user base.
What caused this shift?
Unlike the early internet, which required minimal infrastructure to connect billions, frontier AI relies on highly concentrated, resource-intensive, and significantly expensive computing power. Governments no longer view this advanced software as a bridge for global connection, but as a sovereign weapon. As a result, the world's leading AI powers are locking down their digital borders and sealing off their AI ecosystems, replacing the open web with heavily guarded algorithmic fortresses.
Threat of distillation and intellectual property theft
This reversal raises an obvious question: why are nations locking down technologies that naturally thrive on open collaboration? The answer lies in the intense competitive anxiety surrounding intellectual property. AI models cost billions of dollars and take months of computing power to train, yet they are remarkably vulnerable to digital piracy.
A US House Select Committee on China report titled "Buy What It Can, Steal What It Must" formally said Chinese labs were actively extracting frontier capabilities from US developers via industrial-scale fraud to strengthen their domestic tech stack. Earlier this year, Anthropic claimed that Chinese AI firms, including MiniMax and DeepSeek, utilised over 24,000 fraudulent proxy accounts to execute 16 million queries against the Claude model, according to a report published by the New York Times.
This campaign was designed to facilitate model distillation by leveraging American AI reasoning to train domestic models and bypass significant research costs. Consequently, the US government initiated measures, including National Security and Technology Memorandum 4 (NSTM-4) and proposed legislation to combat industrial-scale intellectual property extraction. Officials argue that adversaries could use millions of automated responses as high-quality training data to teach their own models how to reason and code, bypassing billions of dollars in research and development.
Beijing has officially dismissed the US intelligence claims and the IP theft labels as politically motivated fear-mongering. The prominent voices in the Chinese tech ecosystem argue that US tech giants invoke national security, primarily to protect their corporate monopolies and the premium API subscription revenue against competition from cheaper and highly efficient Asian competitors. From China's perspective, distilling logic from massive, closed-source Western models helps them build lean, high-performing open-weight models like DeepSeek-R1. By making these efficient models accessible globally, Chinese firms argue that they are democratising AI, breaking the Western corporate stranglehold on foundational reasoning tools.
Other than this, the threat of losing the AI arms race has forced the US government to step in with official executive policy, treating AI models with the same gravity as nuclear secrets. In April 2026, the White House Office of Science and Technology Policy issued National Security and Technology Memorandum 4 (NSTM-4). The memo officially designated adversarial AI distillation as a direct threat to US innovation and national security, mobilising federal resources to protect private labs. Following NSTM-4, the US State Department issued a rare, sweeping global warning to foreign allies, formally naming Chinese AI labs and instructing diplomats to caution international partners about the dangers of unchecked model extraction. To legally back this, the US Congress also introduced the Deterring American AI Model Theft Act, which aims to formalise how proprietary model weights, codebases and critical training data are protected by federal security clearances and the strict export bans.
China, too, has begun restricting overseas travel by leading AI scientists, founders and executives. The aim is to prevent technology leaks, stem a brain drain to Western tech giants and safeguard artificial intelligence as a core national security asset. With models like DeepSeek achieving massive breakthroughs in computational efficiency, China considers its optimisation techniques to be highly sensitive corporate and state secrets that shouldn't leave its borders.
To conclude, the paths of these AI powers may diverge, but they lead to the same destination: a deglobalised world. The restrictions on travel for top data scientists highlight that the AI race is no longer fought over software, but over human capital. By treating algorithmic minds as national security assets, China is drawing a firm border around its intellectual property to survive aggressive Western chip curbs. This locking down of talent completes the transformation of the digital world: the open, globalised internet that thrived on borderless collaboration is permanently giving way to isolated, sovereign ecosystems, where the ultimate power belongs to the nations that can successfully control both their computing infrastructure and their talent base.
To navigate this 'Splinternet' and survive an era of AI deglobalisation, middle-powers like India must shift from being mass technology consumers to self-reliant technology creators. Rather than trying to beat the US or China in a trillion-dollar war over massive frontier models, India's strategy should focus on selective AI sovereignty, localised efficiency and leveraging its population-scale data. This also presents India with an opportunity to diplomatically leverage its India AI Mission as a viable model among other AI middle powers.
The author is doctoral scholar, SIU-JRF, Symbiosis School of International Studies, Symbiosis International (Deemed) University, Pune.
