When an AI Company Has to Delete Its Customers to Survive

Eight months after Meta bought Manus for more than $2 billion, the deal is being unwound, users are being told to back up their accounts before they are deleted, and geopolitics has arrived rather abruptly in the inbox.

This morning I received an email from Manus with a subject line designed to get your attention: ACTION REQUIRED. Your account will be deleted.

Not suspended or migrated. Deleted.

I have until 23 August to back up my Manus task data and retain a separate account package. The service will then be unavailable for affected users for two days before Manus resumes independent operations and allows users to restore their data.

It sounds like an unusually dramatic cloud migration. It isn’t. It is the customer-facing consequence of a $2 billion technology deal being put into reverse.

The deal that had to be unhappened

Meta acquired Manus in December 2025. The attraction was obvious, Manus had emerged as one of the more interesting AI agents, designed not merely to answer questions but to carry out multi-stage work involving research, documents, websites and data.

But Manus was founded by Chinese entrepreneurs before moving its centre of operations to Singapore. When one of America’s largest technology companies acquired it, Chinese regulators began examining whether technology and intellectual property originating in China could legally have been transferred.

The result was extraordinary. China effectively ordered the transaction to be reversed.

Buying a technology company is one thing. Unbuying one after systems, people, intellectual property and customer data have begun moving together is considerably harder. Corporate lawyers can reverse ownership structures but disentangling technology is much messier.

Hence the email sitting in my inbox.

Manus says this isn’t a cyberattack or data breach and I’ve got the same message in the app and online. Certain account and task data must be deleted as part of its return to independent operations and to meet regulatory requirements. Users such as me therefore have to back it up ourselves and restore it afterwards. If we can be bothered. I can’t.

Your AI account has a nationality

For most of the internet era, we behaved as though software somehow existed above geography. Cloud computing simply hid the geography rather well.

AI is exposing it again.

Governments increasingly care about where AI intellectual property was created, who owns it, where the data resides and whether strategically important technology can be transferred overseas. These sound like questions for lawyers and governments until they result in an ordinary customer being told his account is going to disappear.

This is therefore about much more than Manus. The global AI contest isn’t simply OpenAI versus Google, Anthropic versus Meta or Claude versus ChatGPT. Underneath the competition between products is an increasingly important competition between jurisdictions.

What does it mean for me?

Fortunately, not very much.

Most of my substantial work with Manus was between January and March, when I was using it in connection with a radio show I was doing. I subsequently became too ill to progress that work, so Manus dropped out of my regular workflow for a while.

Recently I have started using it again for something completely different. I own a domain that has attracted persistent spam, and I have been using Manus rather successfully to track the activity, analyse patterns and isolate the spammers.

It is actually a good example of where AI agents become interesting. Rather than simply answering individual questions, Manus can work through a slightly messy investigation, following information and gradually narrowing down what is happening.

I will therefore back everything up, but fortunately I don’t have a business-critical operation sitting inside Manus. For someone who has spent months building workflows, research or processes inside it, this week’s announcement could be considerably more disruptive.

The wider warning

That is perhaps the other lesson in this story.

We are rapidly entrusting AI platforms with research, documents, strategies, analysis, project histories and increasingly entire workflows. As agents become more capable, the valuable asset may not simply be the document they produce but the accumulated context that allowed them to produce it.

Manus demonstrates why important work should not exist exclusively inside somebody else’s AI platform. Platforms get bought, products disappear, terms change, regulators intervene and companies fail. Occasionally, a superpower orders your AI provider to reverse a $2 billion acquisition.

The answer isn’t to stop using cloud AI. It is simply to remember an old rule in a new technological world – if the work matters, keep a copy you control.

The Lomond Logic

The fascinating thing about the Manus story isn’t whether Meta loses money or who eventually owns the company. It is that an abstract geopolitical battle involving China, America, Singapore and artificial intelligence has produced a very concrete instruction on my laptop:

Back up your files because your account is about to be deleted.

AI is moving from novelty to infrastructure, and the closer it gets to infrastructure, the more governments will care about who owns it, where it lives and who controls it.

But there is a mirror image of that question for the rest of us. While governments worry about who owns the AI companies, perhaps we should pay a little more attention to who controls the work we put inside them.

The first generation of the internet encouraged us to believe technology had abolished borders. Artificial intelligence may be about to remind us that the borders were there all along.

As always, the technology is fascinating. The consequences are rather more interesting.