Manus Goes Independent Again as Meta Unwinds Blocked $2B Deal — User Data Deletion Underway

One of the strangest corporate sagas in recent AI history reached its formal conclusion this week. Manus — the agentic AI startup that became a global sensation in early 2025 — is once again an independent company, after Meta was forced to unwind its $2 billion acquisition on the orders of Chinese regulators. And as part of the separation, a mass deletion of user data began on August 23 and runs through August 24, Singapore time.

For anyone who followed the story from the beginning, the arc is almost cinematic: a Chinese startup goes viral overnight, gets acquired by one of the world’s largest tech companies, and then has the deal ripped apart by regulators four months later — leaving users scrambling to back up their work before the deadline.

What happened

The timeline, reconstructed from Reuters, CNBC, and the Wall Street Journal:

• March 2025: Manus, built by the Chinese startup Butterfly Effect, goes viral as one of the first genuinely capable autonomous AI agents — it could plan trips, analyze stocks, and complete multi-step tasks with minimal supervision.
• Late 2025: Butterfly Effect re-incorporates in Singapore, a move widely read as an attempt to bypass U.S. investment restrictions on Chinese AI firms.
• December 2025: Meta acquires Manus for roughly $2 billion, one of its largest AI talent-and-product acquisitions.
• April 27, 2026: Chinese authorities, citing national security grounds, order Meta to unwind the completed deal. The WSJ reported that Beijing claimed jurisdiction because Beijing Butterfly Effect Technology — the original Chinese entity — remained subject to Chinese law regardless of the Singapore re-incorporation.
• August 11, 2026: Manus announces it will resume operating as an independent company. It also warns that data created by affected users on or after December 29, 2025 will be deleted as part of the separation.
• August 23–24, 2026: The deletion window opens. Users had until 7:59 a.m. SGT on August 23 to export their data; deletion runs through August 24.

Reuters characterized the episode bluntly: a completed cross-border acquisition, ordered undone retroactively. That is exceptionally rare in global M&A, and it is now a precedent.

Why Beijing blocked it

The WSJ’s reporting pointed to national security review under China’s expansion of antitrust and national security vetting of foreign acquisitions involving domestic technology firms. The core of Beijing’s argument: Manus was a Chinese company. Whatever paperwork said otherwise — Singapore holding company, U.S. acquirer — the underlying technology, the founding team, and the original operating entity were Chinese. Allowing an American giant to absorb it would mean transferring Chinese-developed agentic AI capability to a U.S. company at a moment when both governments treat AI as strategic infrastructure.

Fortune’s analysis in April framed the block as a marker of how far Washington and Beijing have drifted apart on AI: the same deal, done two years earlier, likely closes without incident. In 2026, it was politically impossible in both directions — U.S. regulators were already scrutinizing Chinese ties in the restructured deal, and China refused to let the asset leave.

The data deletion, explained

The most concrete consequence for ordinary users is the deletion now underway. Because Manus operated under Meta’s infrastructure during the acquisition period, data created by affected users on or after December 29, 2025 — essentially everything from the Meta era — cannot simply be carried back to the independent company. Deleting it is the cleanest way to sever the data flows between the two businesses.

If you were a Manus user during that window and did not export your data before 7:59 a.m. SGT on August 23, that data is now gone or going. The company has been direct about this: the deletion is part of the separation, not optional, and not reversible.

What it means

For cross-border AI M&A, this is a new red line. A completed deal being unwound by a third-country regulator tells every AI startup with Chinese origins — and every Western acquirer — that re-incorporating in Singapore does not launder away regulatory exposure to Beijing. Expect deal lawyers on both sides to demand much deeper analysis of “residual jurisdiction” before touching any AI company with Chinese roots, however it is structured.

For Manus itself, independence is survivable but harder. The company returns with its product, its team, and its brand — but without Meta’s distribution, compute budget, and capital. It also returns to a market that has moved: agentic AI is now crowded, with offerings from OpenAI, Anthropic, Google, and a wave of well-funded startups. Manus was a pioneer; being a pioneer in 2025 does not guarantee relevance in late 2026.

For users, it is a reminder that AI-generated work is data. Conversations, agent runs, generated files — all of it lives on someone else’s infrastructure, subject to corporate deals and regulatory orders you have no say in. The Manus deletion is the largest example yet of users losing AI-generated work not to a bug or a hack, but to geopolitics. If your agent’s output matters, export it. Regularly.

For the industry, it confirms that AI is now governed by two regulatory poles. U.S. export controls restrict what American AI can flow to China; Chinese security reviews restrict what Chinese AI can flow to America. Companies caught in between — and there are many, given how international AI talent and capital are — will increasingly face an impossible choice: pick a pole, or accept that any deal can be unwound after the fact.

The bottom line

Meta paid $2 billion for Manus, ran it for less than eight months, and is now giving it back — with user data deleted as the price of separation. Manus gets its independence back, but in a far more competitive market and under a regulatory microscope it cannot escape. The Manus saga will be cited for years as the case study in what happens when AI nationalism collides with a globalized AI industry. And for the users whose December-2025-to-August-2026 work vanished this weekend, it is a lesson learned the hard way: in this era, your AI’s memory belongs to whoever the regulators say it belongs to.