When code has a passport: How the China-US AI war sparked a new regulatory tug-of-war

share_windows This article appeared in The Diplomat on January 15 2026. 

As 2026 dawned, Meta announced a $2 billion acquisition of Manus, a fast-rising artificial intelligence firm specialising in agentic AI. What promised to be a seamless union of US platform power and seemingly stateless frontier technology was soon eclipsed by a probe from China’s Ministry of Commerce (MOFCOM). 

Ahead of the deal, Manus had undergone a radical 'identity reconstruction': shifting its headquarters to Singapore, reincorporating as a Singaporean entity, migrating its core team, and hollowing out its Chinese operations. By redomiciling in the city-state, Manus sought to present itself as a stateless firm, scrubbed of its Chinese DNA and ready for the warm embrace of Silicon Valley capital. Such maneuvers have become a survival strategy for AI start-ups seeking global expansion amid intensifying great power competition for technological supremacy. 

Yet the regulatory backlash is revealing. In the AI age, a firm cannot rewrite the history of its technology through 'Singapore washing.' What was intended as a commercial exit has instead become a landmark test case in how compliance, AI sovereignty, and the 'nationality of technology' are reshaping the geography of global innovation.

Corporate mobility, technological immobility

Manus’s parent company, Butterfly Effect, was founded in 2022 by Xiao Hong, a serial entrepreneur. Originally a fully Chinese firm with operations in Wuhan and Beijing, Butterfly Effect was built by a Chinese engineering team. It launched Manus ('mind and hand') – an 'autonomous AI agent' designed for high-value knowledge work – in March 2025 and became an overnight success. By April 2025, Manus had secured a $75 million Series B round led by Benchmark Capital, propelling its valuation to nearly $500 million. To avoid what investors feared could become 'reverse-CFIUS' scrutiny in the United States – and to pursue a multi-billion-dollar exit – the founders concluded that they had to bridge the widening gulf between Chinese innovation and Western capital. This triggered a two-pronged strategy of identity reconstruction.

First came legal restructuring. By mid-2025, a Cayman-based holding company and a Singapore operating hub had replaced the firm’s Beijing-centered structure, shifting operational control from Butterfly Effect Technology in China to Butterfly Effect Pte in Singapore.

Second came physical migration. In June 2025, Manus relocated its headquarters. Of its 120 employees, only 40 core engineers were transferred to the city-state; the remainder were laid off.

After the announcement of the acquisition, Meta spokesperson Andy Stone stated that Manus would retain no Chinese ownership interest and would terminate all operations in the People’s Republic. This 'double divestiture' was designed to scrub the company of its Chinese DNA and present it as a global – or more precisely, a stateless – enterprise.

The strategy rested on an assumption inherited from hyper-globalisation: that companies can move, and so can their assets. But in the AI era, assets are no longer factories or brands; they are data, algorithms, and the talent that builds and trains them.

Unlike traditional large language models (LLMs) that function as passive chatbots, Manus is an 'agentic' system – an autonomous software entity capable of reasoning, planning, and executing complex, multi-step tasks with minimal human intervention. Its architecture relies on what the company calls 'context engineering': a perception layer to gather data, a cognitive layer to set goals, and an execution layer to carry them out. These systems improve continuously as they absorb user behavior at scale, creating a self-reinforcing feedback loop between data and their operating environment.

It is precisely this context-engineering flywheel that regulators may treat as a core asset subject to export control.

Redefining sovereignty in the AI age

What emerges from the Manus case is a sharper conception of technological nationality. Technology assets, regulators are signaling, have national roots. Their identity is determined not by where a holding company is incorporated, but by where code is written, algorithms are trained, and data feedback loops are formed. In other words, the provenance of core algorithms matters.

Regulators in Beijing argue that Manus’ intelligence flywheel is not easily uprooted. The metaphor is botanical rather than digital: a tree may grow tall elsewhere, but if it was nourished by local soil and water, uprooting and selling it raises legal questions about who owns the roots.

The foundational research and development of Manus took place in China. Its roots – the core code – were shaped by the data environments and engineering practices of its Beijing-based team. For the Chinese state, allowing such an asset to be transferred wholesale to a foreign giant like Meta is not seen as a commercial success, but as a 'malicious loss of strategic technology.' In an era in which algorithms have become strategic assets, core code is no longer merely a product – it is part of the national interest.

Legal cosplay

In cross-border technology transfer, the behavioral logic of major powers is converging. Both Washington and Beijing are moving towards a model of 'technological piercing,' reserving the right to intervene in any transaction that threatens their perceived sovereignty over intellectual property. Chinese entrepreneurs may pursue 'de-Sinicisation' to reassure American investors, yet US enforcement practice – including tightened restrictions on outbound investment into China – demonstrates that attempts to escape substantive jurisdiction through formal restructuring are fragile when core national interests are at stake. What Chinese regulators care about is not the identity of a Singaporean shell company, but the process by which technology moves from Chinese territory to foreign control.

Jurisdiction does not rest on a registration certificate; it rests on substantive nexus. Whether it is the United States asserting extraterritorial reach based on technological origin, or China scrutinising a deal based on algorithmic provenance, the logic is the same. Regulators are looking past the Singaporean façade to ask: Who wrote the code? Where was it developed? Whose data trained it?

As the Manus case shows, identity reconstruction can amount to little more than 'legal cosplay.' Its success depends not on elegant corporate architecture, but on whether a firm becomes important enough to trigger state intervention.

Seen this way, Manus’s relocation was not a shield but a lightning rod. By hollowing out its domestic operations to secure a higher valuation abroad, it crossed a strategic red line. 

The regulatory net

The United States has tightened restrictions on advanced-semiconductor exports to China. Beijing responded with stricter technology export controls, deeming them necessary to protect national technological competitiveness. Throughout 2024 and 2025, China’s Ministry of Commerce and Ministry of Science and Technology issued updates to the Catalogue of Technologies Prohibited or Restricted from Export, explicitly covering AI, including interactive interfaces, voice recognition, and personalised information-push technologies based on data analysis.

The introduction of the 'Affiliates Rule' (or '50 percent rule') in September 2025 marked a watershed. It presumes denial of export licences for any entity on China’s control or watch lists and extends that presumption to their global subsidiaries. For Butterfly Effect, which retained a Beijing-based parent until mid-2025, the legal connection to Chinese jurisdiction therefore remains sticky.

MOFCOM’s review seeks to determine whether Manus’ core technology was developed in China and, if so, whether its transfer abroad complied with China’s export-control laws. Even if the technology is not ultimately classified as 'strategically critical,' Beijing is determined to manage the precedent of a blockbuster AI exit to a foreign platform. The message is that global expansion must align with national interests, not proceed through corporate self-amputation.

Singapore’s geopolitical tightrope

Singapore’s rise as a hub for 'Singapore washing' reflects its status as a geopolitical buffer zone. As China-US technology rivalry intensifies, the city-state has become the preferred jurisdiction for firms seeking access to both ecosystems. TikTok and Shein have followed the same path.

But this role carries strategic risk. If Singapore becomes the main conduit through which Chinese-trained AI is repackaged for American platforms, it will face growing regulatory and diplomatic pressure from both sides. What looks like corporate arbitrage is, in reality, regional geopolitics playing out through registries and data pipelines.

The Meta-Manus deal also highlights the growing importance of data governance. Big Tech’s power lies in its ability to harvest, analyse, and monetise continuous data flows that feed self-reinforcing AI systems. These systems entrench market power, and strain national regulatory frameworks. The case underscores the need for global rules not only on privacy and competition, but on data flows and technological sovereignty.

For Singapore, the implications are profound. It is no longer just a financial hub or logistics center; it has become a front line in the battle over the nationality of technology.

The new playbook for cross-border innovation

The real 'butterfly effect' of the Meta-Manus deal is not the transaction itself, but the realisation that in the AI era the decisive question is shifting from which model you use to whether you can operate within a complex geopolitical framework.

For entrepreneurs, the lesson is stark. Legal engineering cannot substitute for substantive compliance. Geopolitics is no longer a 'black swan' risk to be managed at exit; it must be embedded in a firm’s design from the first line of code.

This demands a cognitive upgrade. The 'flat-world' assumption – that technology companies can be stateless citizens of a digital economy – has collapsed. The most resilient firms will be those that respect the sovereignty red lines of all relevant jurisdictions.

In this new era, code does indeed have a passport. It carries the weight of its origins and the scrutiny of its home state. Firms that ignore this reality may find that, however global their ambitions, borders reassert themselves at the most inconvenient moment. The geography of innovation has changed, and those who fail to read the new map will find themselves lost along its compliance fault lines.

As 2026 unfolds, the outcome of the MOFCOM probe will serve as a bellwether for the tech industry. Will it lead to a complete block of the acquisition, a forced divestiture of algorithms, or strict export-licensing conditions? Whatever the outcome, the message is clear: 'Singapore washing' cannot cleanse the genetic code of technology.

The China-US technology competition will continue, but regulated cooperation will remain the dominant pattern. For China, the challenge is to protect strategic intellectual property while remaining part of the global innovation cycle. For the United States, it is to absorb frontier technologies without compromising national security.