Author: Amelia I Biteye Content Team
On April 27, 2026, the Office of the Working Mechanism for Foreign Investment Security Review (National Development and Reform Commission) made a decision to prohibit foreign investment in the Manus project in accordance with laws and regulations, and required the parties to withdraw the acquisition transaction.
In just a few dozen words, the deal worth over $2 billion was abruptly terminated. Manus's years of product development, legal framework construction, financing, and exit strategies all crumbled and went down the drain.
This is the first foreign acquisition in the AI field to be publicly halted since the "Measures for Security Review of Foreign Investment" came into effect in January 2021.
This transaction has a unique aspect: both parties are legally based overseas: Meta is a US company, and Manus has completed its relocation to Singapore and established a holding structure in the Cayman Islands. However, Chinese regulators ultimately decided to prohibit the investment.
The spillover effects of this case include AI companies such as Lunar Dark Side, ByteDance, and Leap Star, which are now facing clearer compliance guidance.
Behind this lies a deeper issue: traditional offshore architecture practices are becoming completely ineffective. Entrepreneurs need to clearly define their compliance path from Day 0.
This article doesn't tell stories, it delivers practical information—what laws and regulations govern this; where is the red line drawn for "washing" style overseas expansion; and how companies should choose their strategies starting today.
First, in accordance with laws and regulations, but what laws and regulations are being followed?
Looking back at the Manus case, initial discussions within the industry largely focused on "what happened"—relocation, separation, and injunctions. However, as details of the case gradually surfaced, the legal community's focus shifted to a more fundamental question: On what grounds could regulators halt this transaction? What laws? What regulations?
The answer lies not in any single law, but in a three-tiered, progressive regulatory logic. These three layers work together to form an inescapable framework for oversight.
First layer: Identifying "Chinese subjects" - the underlying basis for penetrating censorship
This is the legal starting point of the entire case: which company is Manus?
From a legal perspective, the answer seems clear – Manus has completed its relocation of its registration to Singapore, its holding structure is based in the Cayman Islands, and its parent company, Butterfly Effect Pte, is a Singaporean entity. This is also the core legal argument of the Manus team throughout the entire transaction:
"Our main structure has been converted to an overseas architecture."
But the regulator's response was:
Form doesn't matter, substance does.
Jingtian & Gongcheng Law Firm systematically analyzed from a legal perspective why the "externalization of legal shell" failed in the Manus case. The root cause lies in the fact that the core AI assets have an unbreakable substantial connection with Chinese legal jurisdiction in four dimensions:
Team Dimension: The team of engineers who have mastered the underlying core logic has accumulated R&D experience in China for a long time, and their technical capabilities have been trained and developed in China.
In terms of computing power: domestic R&D has created a path dependency in technical interfaces and computing power scheduling, and the core system's architecture is marked with a Chinese label;
Algorithm dimension: The research and training of the core model weights were completed within China, which is the most legally significant "technological source";
Data dimension: The training data accumulated from human feedback reinforcement learning (RLHF) based on massive user interactions is highly concentrated within China.
These four dimensions all point to the same conclusion: Manus's legal form is Singaporean, but its "technical substance" as a company—its origin, core, and foundation—is entirely within China. Based on the principle of "substance over form," from a regulatory perspective, this substantive connection is sufficient to form the basis for thorough investigation—the first cornerstone of all subsequent legal actions.
Therefore, although Xiao Hong founded Butterfly Effect Technology in Beijing in 2022, established a "Cayman Islands-Hong Kong-Beijing" red-chip structure in 2023, and relocated the company to Singapore in 2025, completing the team separation and business isolation, the law does not consider "when the relocation occurred," but rather "where it originated." Any technology assets originating in China do not change their nationality due to a mere registration change.
Second layer: Export restrictions and regulatory circumvention - Legal characterization of "bathing-style" overseas expansion
Once the first layer is established—Manus is deemed a "domestic enterprise" in substance—the second layer of legal logic follows: transferring core assets overseas is itself an export activity. Export activities are subject to export control regulations.
Manus's three-step process, in the eyes of regulators, forms a complete puzzle of "circumventing export controls":
The first step was the transfer of the main entity. The company was moved from China to Singapore, establishing an offshore entity, Butterfly Effect Pte, and setting up a Cayman Islands holding structure. Legally, this completed the first step of "de-Sinicization."
The second step involved the relocation of the team and assets. Nearly two-thirds of the staff in the China region were laid off in a swift move (80 out of 120), with over 40 core technical personnel remaining to relocate to Singapore.
The third step was to separate data from business operations. This included clearing out domestic social media accounts, blocking access from Chinese IP addresses, and terminating cooperation with local partners such as Alibaba's Tongyi Qianwen.
Legally, the technical knowledge, R&D capabilities, and algorithmic experience of core technical personnel taken abroad constitute "technology export" behavior that may be covered by the "Catalogue of Technologies Prohibited and Restricted from Export." Furthermore, according to the "Data Security Law" and the "Measures for Security Assessment of Data Export," the large amount of user interaction data trained before the data was cut off originated primarily from within China—the data's genetic code was already written into the model, making the data cut-off process untraceable and impossible to delete.
Therefore, the logic of regulatory penetration can be summarized in a cold, hard statement:
The code is written on Chinese soil, and the data resides among Chinese users – this is “Chinese assets.” Transferring them is exporting them, and exports are subject to regulation.
The essence of "going global in a bath-like manner" is to cover up substantive violations with formal compliance, which is a systematic circumvention of the export control system.
The third layer: proactive reporting mechanism - you can't say "I don't know".
If the first two levels are "substantive violations", the third level is "procedural violations" - and it is the easiest to be convicted of.
Article 4 of the Measures for Security Review of Foreign Investment clearly stipulates that for foreign investments involving important information technology, key technologies, and other fields, the parties concerned "shall proactively declare to the working mechanism office before implementing the investment." This is a mandatory prior declaration obligation, not a "recommendation" or a "reporting after an incident."
Throughout the entire transaction, from start to finish, Manus and Meta never made any proactive disclosure to Chinese regulatory authorities. During the months-long settlement period, Manus and its investors seemed to have reached a dangerous tacit understanding: they wouldn't proactively open the window unless the regulators knocked.
In legal practice, "failure to report a violation" is itself a serious violation. It sends the message that either the violation was knowingly committed or intentionally evaded regulations. In either case, regulators will not let it go unpunished.
A compliance lawyer summarized the situation after the incident as follows:
"The biggest compliance flaw exposed by the Manus case is not the controversial applicability of a particular regulation, but that the company has completely abandoned its reporting obligations to Chinese regulators. In the legal system, evading the procedure itself is more intolerable to regulators than substantive violations."
In hindsight, Manus's fate was actually sealed in the first layer: once the thorough investigation determined you to be a "substantial Chinese entity," the second layer of export control logic and the third layer of reporting obligations automatically unlocked. These three layers of legal reasoning, progressively intricate and interconnected, form a logical closed loop. Within this loop, there is no room for "luck."
II. Why the National Development and Reform Commission (NDRC)?
The Ministry of Commerce took the lead. On January 8, 2026, a spokesperson for the Ministry of Commerce publicly stated that it would conduct an assessment and investigation into the acquisition's consistency with relevant laws and regulations concerning export controls, technology import and export, and overseas investment. However, on April 27, it was the National Development and Reform Commission that made the final decision.
There's more to this departmental switch than meets the eye. Some experts believe the Ministry of Commerce is basing its decision on the "Catalogue of Technologies Prohibited and Restricted from Export," which provides very specific descriptions of controlled technologies: AI user interface technology specifically designed for Chinese and minority languages. However, after its restructuring, Manus has switched all its services to English, effectively excluding Chinese users. This suggests that simply following the export control route could lead to some controversy.
This is the area of contention regarding the applicability of regulations. However, we tend to favor a deeper meaning, since the applicability of laws is less important than political considerations in terms of priority.
The National Development and Reform Commission (NDRC) is in charge of "security review," while the Ministry of Commerce is in charge of "technology import and export." The NDRC's involvement signifies that this matter has transformed from a "business" issue into a matter of "sovereignty."
In other words, the intervention of the National Development and Reform Commission (NDRC), a macro-level department with more comprehensive economic management authority than the Ministry of Commerce, sends a clear signal: this is not an accidental enforcement action against a particular company, but a systemic deterrent aimed at "striking first to avoid being struck a hundred times."
Killing one is to warn the hundred.
All practitioners who were still observing now see where the red line is drawn – not in the ambiguity of a specific clause, but on the indisputable ultimate standard of safeguarding national security.
Three or four high-risk trigger points
Based on the Manus case and the "penetrating review" principle established by the "Measures for Security Review of Foreign Investment," the following four red lines are clear. If any of these lines are crossed, the "shower-style" overseas expansion strategy is no longer viable.
Red Line 1: The founder holds a Chinese passport and has not revoked his Chinese citizenship.
Manus founder Xiao Hong is a Chinese citizen. China's export control law covers natural persons. This means that the founder himself may also become a subject of regulatory scrutiny, and the relevant arrangements cannot be understood solely at the company level.
A harsher reality lies across the Pacific: in North American VCs' geopolitical risk assessments, the financing environment for Chinese founders is also tightening. Leading Silicon Valley VCs like a16z are experiencing a sharp decline in their willingness to invest in founders holding Chinese passports due to geopolitical pressures.
Manus's Series B funding round was led by Benchmark, but Benchmark subsequently faced strong backlash from the US political establishment for the investment, with several Republican senators calling the deal "assisting the Chinese government."
Investors at Silicon Valley Founders Fund were blunt:
The founder is Chinese, the company is based in Beijing, and its core technology is a general AI agent – this is the "original sin".
Both sides are closing their doors. If you have a Chinese passport, American capital is uneasy; if you have Chinese technology, Chinese regulators won't let go. This gap is much narrower than most people imagine.
Red Line Two: Having received money from state-owned assets
It's not only "direct investment by sovereign wealth funds" that counts as state-owned assets. Government-guided funds at all levels, state-owned components in RMB fund LPs, and policy bank loans—these all fall within the scope of "state-owned asset infusion." And those meager subsidies like office space, computing power, and talent allowances—the ones that people complain about during the application process being cumbersome and meager—will all be noted down when it comes to reckoning.
Red Line Three: The first line of code must be written within China.
The initial location where the core code was written, the location where the algorithm model was trained, and the storage location of the technical documentation—these seemingly "purely technical" facts all constitute legal proof of the "source of technology." Manus's early development was completed in China, and when the team moved to Singapore, the code they took with them already constituted a technology export. Manus never made any technology export declaration for this transfer.
Red Line Four: Use of Chinese Data
This is a common illusion among many AI entrepreneurs: that as long as they clear out domestic users and block Chinese IPs later, their company will be clean.
However, in the eyes of regulators, the 'technical substance' is not only about the code, but also about the data structure.
The Data Security Law and the Measures for Security Assessment of Cross-border Data Transfer have clear review requirements for cross-border transfers involving "important data." Although Manus shut down its Chinese service and blocked Chinese IPs, the user interaction data accumulated in the early stages had already completed the core training of the model—the data genes were engraved in the model's weights, and could not be recovered or deleted through "post-processing cleaning." Since the data originated from Chinese users, the model carries a Chinese label.
IV. Entrepreneurs in Specific Industries: Choosing Sides Starts Now
The "Security Review Measures" establish a security review mechanism for foreign investment that may affect national security, focusing on defense and security sectors such as the military industry, as well as important areas where foreign capital gains actual control, such as important information technology, key technologies, major infrastructure, and important resources.
In the current regulatory environment following the Manus case, the following points deserve special attention:
First, the determination of "actual control" in practice is not solely based on the shareholding ratio. If a foreign investor can exert significant influence over a company's operational decisions, personnel, finances, or technology (e.g., possessing veto power or the right to know about key technologies), it falls into this category. This definition is very broad. For example, even if you only hold 5% of the equity corresponding to a US dollar fund, the veto power attached to that 5% equity may be considered "significantly influencing the company's operational decisions," thus qualifying as "actual control" and triggering an investigation.
Second, the National Development and Reform Commission (NDRC), as the leading department in the working mechanism, has the authority to provide compliance guidance based on national security judgments. For example, the NDRC's guidance on April 24, 2026, requiring some AI companies to reject US investment, although not explicitly stated in the regulations, falls under the scope of "routine security review work and preventive management" authorized by Articles 3 and 7 of the "Security Review Measures".
Third, it is not recommended to circumvent scrutiny through VIEs, nominee shareholding, trusts, or other similar methods. In practice, once an arrangement to circumvent scrutiny is identified, the company may face the risk of rectification, suspension, withdrawal, or other compliance measures.
Conclusion: The gray area of "playing both sides" has been completely blocked from all directions. From now on, companies must clearly define their compliance stance from Day 0.
Especially in the AI field, you can only choose one of the following two routes.
Route A: Take the American-funded route - Leave with nothing.
If you decide to take a dollar fund, follow the Silicon Valley route, and your ultimate goal is to be acquired or listed on the US stock market, then what you need to do is not to "take a bath," but to completely overhaul your business.
One strict standard: You cannot cross any of the aforementioned four red lines.
Specifically, this means four things:
First, the founder needs to address their nationality. A Chinese passport is inherently seen as a compliance risk by US venture capitalists. If you're determined to go down this path, giving up your Chinese citizenship is not an option, but a prerequisite.
Second, do not use state-owned funds. All funds involving government-guided funds, state-owned LPs, and policy-based loans should undergo thorough compliance due diligence at the initial financing stage, and should be liquidated or repurchased if necessary.
Third, the source code must be located overseas. This is the most brutal and crucial point. The very first line of code for the core algorithm must be completed overseas. Domestic teams can only work on non-core modules or peripheral businesses. You need to establish a truly capable overseas technology center from the very beginning—not just a shell, but a physical entity.
Fourth, data and users should be kept separate from day one. Never touch Chinese user data from the beginning. It's not about "post-processing," but about "never owning" it in the first place.
The prerequisite for taking this path is that you can bear the cost of a complete break from the domestic market. You'd have to give up all revenue, users, and brand synergy in the Chinese market. You're betting that the returns from globalization will cover this cost. Moreover, even if you achieve all of the above, you'll still face an increasingly unfriendly America – the founder's Chinese identity remains a "sin" in the eyes of certain forces in Silicon Valley.
Route B: Take the domestic investment route - align with the national team.
If you don't want to, or can't, follow the American capital route, then make compliance your moat.
The core logic: China's land and soil can only grow Chinese RMB.
First, proactively embrace state-owned and private capital. In financing, prioritize RMB funds, government-guided funds, and central state-owned enterprise investment platforms. This is not a forced choice, but a strategic alignment: state-owned background is the strongest regulatory pass.
Second, make compliance a first-mover advantage. While your competitors are still trying to circumvent regulations, you proactively apply for security reviews, proactively complete data classification and grading, and proactively register technology exports. In the eyes of regulators, you are "one of us"; in the eyes of the market, your investment in compliance is a barrier that latecomers cannot catch up with in the short term.
Third, turn certification into licensing barriers. Certifications for domestic IT innovation, data security capability maturity certification, and "specialized, refined, and innovative" designations in related technical fields—these aren't costs, they're licenses. In a tightening regulatory environment, having a license versus not having one is a matter of life and death.
Fourth, proactively declare security reviews. According to Article 4 of the "Measures for Security Review of Foreign Investment," foreign investments involving important information technology and key technologies must be proactively declared before implementation. For state-owned enterprises, this is not a burden, but rather the best way to demonstrate your stance to regulators.
By taking this path, you accept the valuation logic and exit pace of RMB funds – a quick $2 billion flash acquisition may not apply to you, but what you gain is stable policy expectations and the right to continue operating in the domestic market.
There is no third way to grow big.
The "Cayman holding company + Singapore operations + domestic R&D + USD financing" fence-sitting model is already doomed. Continuing to hesitate on this path isn't flexibility, it's danger. Regulators won't grant you an exemption just because you haven't thought things through.
If you choose American capital, leave cleanly. If you choose domestic capital, be completely bound to it.
This is the only operational manual that the Manus case leaves for AI cross-border entrepreneurs.
In conclusion: The butterfly effect—a self-fulfilling prophecy.
Manus named its parent company Butterfly Effect. Looking back at that name now, one can only sigh at how prophetic it has become.
This butterfly flapped its wings twice, stirring up two storms. One was an acquisition offer from Silicon Valley, and the other was a ban from Beijing. Now, the regulatory pressure from both sides has taken shape, turning the acquisition offer into a compliance bubble. This case will be written into the financing memos of every cross-border technology company thereafter.
Looking back at that seemingly perfect path of "monetizing and exiting in 9 months, followed by a $2 billion acquisition," it actually concealed three major pitfalls from the very beginning:
Technological minefield: The moment the core AI code is generated within China, it is subject to regulation when it is shown on camera;
Data minefield: Once data from China has been used, it cannot be traced back.
Identity minefield: In this era, technology has a nationality, and so do the people who work with technology.
Acting in accordance with laws and regulations used to be a principle, and now it is an ironclad rule.
The focus today is not on convicting anyone, but on recognizing a trend: the gray areas that were previously used to maneuver by changing the place of registration, structure, and entity are being continuously compressed. For founders, going global is no longer a game of "circumventing regulations first and then fulfilling compliance requirements," but rather a game where they need to figure out their entity, funding, technology, data, and application procedures from Day 0.
I hope that every founding team that is finding a way out in the cracks of the times, whether you choose to go all out on the track of American capital or to cultivate the system of domestic capital, can see the rules clearly, stand firm, and go further.
*This article represents a subjective analysis by our editorial team based on publicly available information and industry observations, aiming to provide a multi-dimensional perspective for discussion. Nothing in this article constitutes legal advice or investment advice. For specific legal issues or business decisions, please consult a licensed lawyer.
