Key Takeaways
- China’s National Development and Reform Commission (NDRC) blocked Meta’s $2 billion acquisition of autonomous AI agent developer Manus AI on April 27, 2026, citing national security and export control concerns.
- The NDRC’s intervention came despite Manus AI’s parent company relocating its headquarters from China to Singapore in July 2025, with Chinese regulators reportedly barring the startup’s co-founders from leaving the country in March 2026.
- If no new acquirer emerges, Manus AI faces a genuine risk of ceasing operations, setting a precedent that could deter foreign investment in Chinese-origin AI startups even after they have restructured offshore. China has blocked one of the most closely watched AI acquisitions of the year, and the consequences extend well beyond the two companies involved. The NDRC’s April 27 decision to reject Meta‘s $2 billion bid for Manus AI signals that Beijing is prepared to assert jurisdiction over AI assets even when the companies behind them have deliberately restructured to reduce their Chinese footprint. For the broader AI investment community, the ruling raises an uncomfortable question: how far does Chinese regulatory reach extend into offshore AI ventures?
China Blocks Meta’s $2 Billion Manus AI Acquisition
Manus AI was developed by the Chinese startup Butterfly Effect, also known as Monica, which relocated its headquarters to Singapore in July 2025 and reduced its China-based team, a move widely interpreted as an attempt to distance the company from Chinese regulatory oversight ahead of a foreign acquisition. It did not work. Chinese regulators asserted jurisdiction anyway, and according to reports, co-founders Xiao Hong and Ji Yichao were barred from leaving the country in March 2026 before the NDRC formally blocked the deal the following month.
Meta had announced its intent to acquire Manus in December 2025, making the April 27 ruling the end of a months-long review process. The NDRC cited national security risks and export control concerns, the same regulatory logic Beijing has applied to semiconductor and telecoms deals in recent years, now extended explicitly to AI agents. The decision suggests that Chinese-origin intellectual property and talent carry a regulatory claim regardless of where a company is formally incorporated. For legal and compliance teams advising on cross-border AI transactions, that is a significant shift in the operating assumption.
The practical fallout is substantial. Meta is expected to return its $2 billion payment, and Manus’s products and data are anticipated to be removed from Meta’s systems. According to reports, there is a genuine possibility that Manus AI could cease operations entirely, as the regulatory overhang may deter new acquirers or investors. A return to the Chinese domestic market, meanwhile, would place the company in direct competition with well-resourced local players in an already crowded field. This ruling could function as a chilling precedent for other Chinese AI startups pursuing international capital, adding compliance complexity to any deal involving Chinese-origin AI assets, intellectual property or talent. For a broader look at how AI regulation is evolving on both sides of the Atlantic, see our coverage of the EU AI Act vs. NIST RMF.
Manus AI’s Core Capabilities: Automating Complex Workflows
Launched in March 2025, Manus AI positions itself as a fully autonomous general AI agent, designed to bridge the gap between human intention and execution. Unlike traditional chatbots that require continuous step-by-step prompting, Manus is engineered to plan and execute multi-step tasks independently from a single high-level instruction. That capability rests on a multi-agent architecture in which a central executor agent orchestrates specialised sub-agents handling planning, knowledge retrieval, code generation and verification.
The platform’s claimed versatility spans a wide range of business applications. It can conduct resume screening and candidate ranking, produce financial analysis reports, build functional websites and automate the creation of personalised content such as travel itineraries. It integrates with external tools including web browsers, spreadsheets, databases and platforms such as Gmail, Google Calendar and Google Drive. A “My Computer” feature reportedly allows direct access to local files and applications on a user’s machine. Tasks run asynchronously in cloud-based virtual machines, meaning the system continues working even when the user is offline.
A May 2026 update introduced a “Make a Copy” feature enabling instant duplication of AI-built websites, giving users a way to experiment without overwriting production assets. For developers and agencies building client-facing products on the platform, that kind of version control is a meaningful addition. The company also allows users to define and reuse “skills”, customised AI capabilities for specific domains such as legal review or financial analysis, which can be combined into multi-step automated workflows rather than one-off task completions.
Where Manus AI Falls Short: Technical Hurdles and Market Realities
The platform’s limitations are well-documented, and several are structurally significant rather than merely cosmetic. Manus struggles with pixel-precise graphical user interface interactions because it lacks native computer vision capabilities to dynamically interpret screen elements. It relies instead on predefined workflows, which limits its usefulness in tasks requiring precise visual manipulation, such as navigating design software or testing responsive layouts.
Early beta users reported frequent server availability problems, with the system often returning “busy” errors during periods of high demand, along with bugs including empty ZIP file outputs. A stable release, Manus 1.6, issued in December 2025, and subsequent updates have addressed some of these issues, but initial instability damaged user confidence. Restrictive context lengths also remained a recurring complaint, limiting the system’s ability to maintain coherence across complex, extended projects.
The pricing model introduces its own friction. Manus operates on a credit-based system, where intensive tasks such as deep SEO research can consume a significant number of credits per session, and the Manus Pro tier is priced at around $199 per month. That contrasts with the flat-rate subscriptions offered by several competitors and makes cost forecasting difficult for enterprise teams running high-volume workflows.
For sensitive or regulated use cases, the architecture raises additional concerns. The system lacks the granular controls and mandatory compliance checks that medical coding, legal review or financial services workflows typically require. Because Manus is reported to use large language models as its underlying reasoning layer, including, in some versions, Anthropic’s Claude, it inherits the hallucination and bias risks common to LLM-based systems. The lack of documented process-level isolation between the model and user credentials has prompted some practitioners to recommend self-hosted or local-first alternatives for sensitive data environments.
Competition is also intensifying. Devin AI targets software engineering autonomy; Perplexity is faster and cheaper for research tasks; and platforms such as Replit Agent focus on app building with integrated backends. Manus’s general-purpose positioning, once a differentiator, is increasingly challenged by specialised agents that perform narrower tasks more reliably. The acquisition block compounds this pressure: without Meta’s distribution and infrastructure, Manus’s path to the scale needed to compete globally is considerably less clear. This dynamic mirrors broader concerns about how AI models handle evaluation and deployment, which our coverage of the Mpathic mPACT benchmark examines in detail.
The Geopolitical Headwinds and Future Outlook
The NDRC’s ruling does more than close off one acquisition. It establishes a regulatory principle: that Chinese-origin AI assets retain a national security dimension regardless of subsequent corporate restructuring. Butterfly Effect’s Singapore relocation was a deliberate attempt to reduce that exposure, and Beijing rejected it. That outcome will be studied closely by every Chinese AI startup currently holding offshore capital or pursuing international buyers.
The implications for cross-border AI dealmaking are substantial. Legal advisers will now need to assess not just where a target company is incorporated, but where its intellectual property was developed, where its key personnel were trained and whether its core technology was built under Chinese regulatory jurisdiction. Export control analysis, already a standard part of US-side due diligence for Chinese tech acquisitions, will likely need to account for the possibility of reciprocal Chinese intervention, something deal teams had not historically priced in.
For Manus AI specifically, the near-term outlook is uncertain. Without a new strategic partner willing to navigate the regulatory exposure, the company may struggle to fund continued development at the pace its competitive position demands. Its multi-agent architecture and autonomous task execution capabilities represent a credible technical foundation, but technical merit alone will not resolve the jurisdictional overhang that now defines the company’s situation. How Beijing handles the next comparable case will determine whether this ruling marks a one-off intervention or the start of a systematic approach to controlling outbound AI. For more coverage of AI policy and regulation, visit our AI Policy & Regulation section.
Originally published at https://autonainews.com/china-vetoes-metas-2b-manus-ai-deal/
Top comments (0)