Manus Returns to Independent Operations as Meta's $2B Acquisition Fully Unwinds

A Deal Unraveled by Geopolitics

On August 11, 2026, Manus—the autonomous AI agent that became one of the most talked-about products of the generative AI boom—formally announced it will resume operating as an independent company, closing the book on one of the most consequential and short-lived acquisitions in AI history. The move comes months after the Chinese government ordered Meta to unwind its roughly $2 billion purchase of the startup, forcing a messy operational separation that has now reached its final, user-facing phase.

For Manus’s users, the announcement arrived with a hard deadline: data generated on or after December 29, 2025 will be permanently deleted between August 23 and August 24, 2026 (Singapore time). Affected users must complete backups before 7:59 a.m. SGT on August 23, with data restoration available starting at 8:00 a.m. on August 25. The clock is ticking for anyone who built workflows, reports, or automations inside Manus during its brief time under Meta’s roof.

How We Got Here: A Nine-Month Saga

Manus was founded in Beijing in 2022 by Xiao Hong (肖弘), alongside co-founders Yichao “Peak” Ji and Tao Zhang, under a parent company called Butterfly Effect (蝴蝶效應). The product launched publicly in March 2025 as a general-purpose autonomous agent capable of independently executing complex, multi-step tasks—browsing the web, writing code, analyzing data, and completing workflows inside a persistent cloud-based virtual computer environment. It was not a chatbot that suggested answers; it was an agent that delivered finished work.

The product went viral. At its peak, Manus claimed a waitlist of over two million people and an annual revenue run rate approaching $90 million. It stood out in a crowded agent landscape by wrapping foundation models—primarily Anthropic’s Claude 3.5/3.7 and Alibaba’s Qwen—inside an orchestration layer that could decompose ambiguous instructions into executable subtasks, run them in a sandboxed environment, and return completed deliverables.

That buzz attracted Meta. In late December 2025, Meta announced it would acquire Manus for more than $2 billion—its third-largest acquisition ever. The deal closed on January 8, 2026, and Butterfly Effect’s three co-founders had already relocated to Singapore in mid-2025, with a Singapore-based entity taking over operation of the Manus product. For a moment, it looked like Meta had secured a flagship agent product to anchor its consumer AI strategy.

Then came Beijing.

China’s Unprecedented Veto

In January 2026, Chinese authorities launched a regulatory review of the transaction. On April 27, 2026, China formally blocked the acquisition using its foreign investment security review mechanism—the same regulatory toolkit the government uses to vet deals that touch national security, critical infrastructure, or sensitive technology. The verdict was blunt: the deal could not proceed. Manus had Chinese roots, had developed technology inside China, and Beijing determined that the transfer of that technology and associated data to a US company constituted a security concern.

The move was unprecedented in the AI sector. While the United States has spent years using the Committee on Foreign Investment in the United States (CFIUS) to block or force divestment of Chinese-linked companies—most famously with TikTok—China had rarely deployed its own investment review powers against a deal involving a Western tech giant. The Manus case signaled that Beijing was willing and able to reach across borders to unwind transactions involving AI technology it considered strategically important, even when the target company had formally re-domiciled abroad.

By June 2026, Meta had completed an operational split from Manus, severed all data-sharing pipelines, and barred Manus staff from accessing Meta’s internal systems. Bloomberg reported that Meta cut off Manus and its employees from the US company’s data infrastructure at the start of that month. The unwinding was underway—but the question of what would happen to Manus itself, and to the users who had built workflows on top of it, remained unresolved until this week.

What Independence Means for Manus

Manus’s return to independent operations is both a relief and a warning. On one hand, the company survives. The agent product, the team, and the technology remain intact under the Butterfly Effect umbrella. Users will be able to restore their backed-up data starting August 25, and the product will continue to operate for businesses and individuals who rely on autonomous task execution.

On the other hand, the episode exposes deep structural risks in the global AI agent market. Manus built its product on top of foundation models from Anthropic and Alibaba—models whose availability, pricing, and terms of service can shift overnight. It operated inside Meta’s infrastructure for seven months, only to be surgically extracted when geopolitics intervened. And it now faces the prospect of a bifurcated future: a potential Chinese joint venture and Hong Kong listing on one side, and an international business operating from Singapore on the other, with the regulatory line between them drawn by governments rather than by the company itself.

The user data deletion timeline is the sharpest illustration of that risk. Anyone who entrusted Manus with sensitive business workflows during the Meta period now has roughly twelve days to export their data before it is permanently destroyed. The company has framed this as a necessary step in completing the regulatory separation, but for users, it is a reminder that the persistence of AI-generated work depends on corporate and geopolitical forces far beyond any single product roadmap.

The Bigger Picture: Agents as Strategic Assets

The Manus saga matters beyond one company. It is the clearest evidence yet that AI agents—the category of products that actually execute tasks rather than just generate text—have crossed the threshold from consumer novelty to strategic asset. Governments are treating them that way. China’s security review did not focus on Manus’s user base or revenue; it focused on the technology itself and the implications of its transfer. The United States has taken a parallel posture, with the Trump administration’s 2026 AI policy framework emphasizing frontier model control and domestic capability retention.

For the AI industry, the lesson is that cross-border agent deals are now subject to the same kind of national security scrutiny traditionally reserved for semiconductors, telecommunications equipment, and defense technology. A startup can build a viral product, sell itself to the highest bidder, and find the transaction unwound by a government that was never at the negotiating table. Founders building agent companies with international ambitions will need to treat regulatory geography as a first-class design constraint—where the team sits, where the data lives, and which models power the orchestration layer are all now questions with geopolitical answers.

Manus’s return to independence is, in that sense, a small victory wrapped in a large cautionary tale. The product lives on. But the era in which an AI agent could be bought and sold freely across borders—like any other software company—appears to be over.