A Deal Comes Full Circle
Manus, the AI-agent startup Meta bought for roughly $2 billion in a deal that closed on December 29, 2025, confirmed on August 11-12, 2026 that it is independent again. Tencent, ZhenFund and HSG led a group of investors who repurchased the company from Meta for the same $2 billion figure. Benchmark, previously Manus's largest outside shareholder, did not participate in the buyback; Tencent bought out Benchmark's stake instead, making Tencent the company's largest shareholder (Caixin Global; CNBC). Manus separately told users it will delete account data generated on or after the December 29, 2025 acquisition date, framing the deletion as a regulatory-compliance step tied to the separation (Manus).
The Order Behind the Reversal
The transaction wasn't undone by market forces. On April 27, 2026, the Office of the Foreign Investment Security Review Working Mechanism — housed inside China's National Development and Reform Commission — issued a decision prohibiting the foreign acquisition of Manus and directing the parties to cancel it, per the agency's own government-disclosure filing (NDRC). The legal basis is the Measures for Security Review of Foreign Investment, jointly issued by the NDRC and the Ministry of Commerce as Order No. 37 of 2020 and in force since January 18, 2021 (gov.cn). This is the first publicly disclosed case in which that review mechanism has been used to compel the unwind of an already-completed AI acquisition — not merely to block a pending one.
What makes the order notable is its target. Manus was founded in China in 2022 but relocated its headquarters to Singapore in 2025, in part to distance itself from Chinese-origin scrutiny as it sought international investors. Singapore's national development minister, Chee Hong Tat, said at the Singapore-China Forum on July 7, 2026 that Beijing's intervention did not violate Singapore law and that Singapore "respects the national security considerations of both China and the U.S." (Caixin Global). That is a statement of non-interference, not an endorsement of Beijing's jurisdictional theory — and the distinction matters, because the company sits outside China's territory and corporate registry.
The Case for a Review — Fairly Stated
Before dismissing this as overreach, it's worth stating Beijing's strongest argument plainly. Every major economy now screens foreign control over AI systems trained on sensitive data or built by teams with access to strategically important know-how — the U.S. does it through CFIUS, the EU through its FDI screening regulation, India through Press Note 3. Manus's core algorithms and engineering team originated in China, and its models were trained in part on data tied to Chinese users. A government that treats agentic AI as a security-relevant technology has a coherent basis for wanting visibility into who ends up controlling it, even after the company reincorporates elsewhere. In that narrow sense, China invoking a five-year-old, previously untested review statute is not exceptional as a category of regulation — CFIUS itself reaches deals where a target's U.S.-linked technology, data or operations create a nexus, regardless of where the acquirer is based.
Where the Comparison Breaks Down
The difference is what triggers the nexus. CFIUS anchors its jurisdiction to control of a U.S. business or U.S.-situated assets — a test tied to where the target operates. The NDRC's Manus order instead reaches a company incorporated, headquartered and (by the time of the deal) operating outside China, based substantially on the Chinese national origin of its founders, engineers and historical user data. That is a materially broader and more portable theory of jurisdiction: it says ancestry of people and data, not location of the business, is what triggers review — years after the company relocated specifically to place itself outside Chinese jurisdiction.
An Outcome That Looks Like Industrial Policy
The practical result of the "security" review is that a frontier AI-agent company, once headed for U.S. ownership, is now majority-controlled by Tencent — a domestic champion — instead of Benchmark or Meta. Regulators are entitled to screen deals on security grounds without their motives being above scrutiny; when the remedy for an alleged security risk consistently reroutes ownership of strategic technology into domestic hands, it is fair to ask how much of the exercise is security review and how much is industrial policy wearing its clothing.
The Chilling Effect Ahead
A published, prospective, incorporation-based test would let founders and acquirers price this risk in advance. A one-line order compelling disgorgement of a deal that closed eight months earlier, applied to a foreign-domiciled company on the basis of its people's origin, does the opposite: it tells any founder with Chinese-origin engineering talent or historical Chinese user data — regardless of where they've since incorporated — that a completed exit to a U.S. buyer can be reversed after the fact. That raises the cost of capital for exactly the kind of cross-border AI entrepreneurship China says it wants to encourage, and it hands Washington's own techno-nationalists a ready-made justification for reciprocal screening of any startup with a Chinese-origin founder. Proportionate security review is legitimate; retroactive jurisdiction keyed to bloodline rather than borders is not, and Beijing would serve its own stated ambitions better by writing the former into law instead of practicing the latter by decree.