I used to think that the greatest threat to AI autonomy was technical centralization—a single model, a single API key, a single point of failure. Then I read the Financial Times report on Manus, the Chinese AI agent startup that was almost swallowed by Meta for $20 billion. The story is not just about a regulatory block or a founder’s travel ban lifted. It is a parable about the fragility of agency in a world where the real custody of power lies not in code but in boardrooms, state security reviews, and equity cap tables.
Here is what the charts won’t tell you: Manus’s founder, Xiao Hong, was grounded in China for months while regulators investigated Meta’s acquisition. Now the ban is lifted, and he is heading back to Singapore. The company will continue independently, with Tencent as the largest shareholder—but not controlling. The deal that could have merged one of the most promising AI agent platforms into the American social media empire was killed by a sovereign state. On the surface, this looks like a victory for independence. But scratch the code, and you will find the same old story: centralized power, even when it protects, still cages.
Context: The Manus Precedent
Manus is a general-purpose AI agent, a product that chains together large language models, tool calling, and multi-step task planning to execute complex workflows autonomously. In 2024, Meta offered roughly $20 billion to acquire the company. The deal was structured to transfer the entire team, technology, and data to Menlo Park. But Chinese regulators intervened, citing national security and data sovereignty concerns. They ordered Meta to withdraw. Xiao Hong and his co-founder, Ji Yichao, were kept inside the country for investigation. Now, months later, the travel restrictions are lifted, and the company is reorganizing: Tencent bought the stake held by Benchmark, a Silicon Valley venture firm, becoming the largest shareholder at under 50%. Manus will continue as an independent entity based in Singapore.
For the crypto world, this should sound familiar. It is the same tension that plays out in every DAO fork, every Layer 2 exit, every regulatory crackdown on DeFi. The question is not whether you are autonomous, but who holds the keys to your survival. In Manus’s case, the keys were held by a handful of regulators in Beijing. They decided that the company should not be absorbed by a U.S. giant. But they did not give the keys to the users or the community. They handed them to another centralized entity: Tencent.
Core: The Technical and Values Analysis
Let me be clear about what this means for anyone building AI agents today. I have spent the last decade dissecting smart contracts, auditing multi-sig wallets, and watching the illusion of decentralization shatter when governance tokens are controlled by a few whales. Manus’s story is a textbook case of the same failure mode, applied to AI.
The first insight is about governance. When a company is saved by a regulator, it is not saved; it is transferred. The regulatory block on Meta’s acquisition was a form of sovereign veto. It prevented a single American corporation from owning the agent. But it did not create a decentralized structure. Instead, it created a new dependency on Tencent, a Chinese tech giant with its own incentives. The fact that Tencent’s stake is capped at 50% is a governance design that mirrors the multi-sig arrangements I have seen in DAOs. But in a DAO, the multi-sig signers are chosen by the community. Here, they are chosen by the state and the largest shareholder. The difference is not trivial. It is the difference between a protocol that can be forked and a company that can be acquired.
The second insight is about data sovereignty. Manus’s agent platform likely ingests vast amounts of user data to improve its task planning and tool calling. Under Meta, that data would have flowed to the U.S. Under the new arrangement, it will likely flow to Singapore and China. The regulatory block was explicitly about data security. But the solution—moving the headquarters to Singapore while keeping the Chinese entity—creates a dual data regime. This is the same problem that crypto projects face when they try to comply with both OFAC and Chinese regulations. The result is a fragmented system where the agent’s behavior changes based on the user’s IP address. That is not autonomy. That is geographic discrimination.
The third insight is about the fallacy of independence. Manus is now “independent,” but its independence is conditional. It depends on Tencent’s continued support, on the regulatory environment in Singapore, and on the goodwill of the Chinese government. In crypto, we have learned that true independence is not about jurisdiction; it is about the ability to verify and enforce rules without a trusted third party. That is why we build on blockchains. That is why we write smart contracts. Manus, despite its technical sophistication, is still a centralized entity. Its code may be elegant, but its fate is decided by humans with pens and passports.
From my experience auditing the Gnosis Safe multi-sig in 2017, I learned that even the most well-intentioned governance can be subverted if the signers are not truly independent. I found 12 critical logic flaws in their implementation—not because the code was bad, but because the assumptions about trust were wrong. The same applies here. The assumption that a regulator or a corporate parent will act in the best interest of the users is a flaw in the architecture of power.
The Human Cost of Corporate Custody
In 2020, during DeFi Summer, I watched friends lose their savings when Compound’s governance token crashed. I interviewed 30 of them. What I heard was not about impermanent loss; it was about the illusion of control. They thought they were participating in a decentralized system, but the system’s fate was tied to a handful of founders and VCs. Manus’s users are in a similar position. They might love the agent’s ability to automate their workflows, but they have no say in who owns the infrastructure. They are renters in a building that can be sold at any time.
The 2021 NFT bubble showed me that authenticity is rare. I refused to mint speculative PFP projects and instead built “On-Chain Diaries,” a small collection of 50 digital artifacts tied to real events in Beijing. I coded the smart contract myself to ensure royalties went to local artists. That project was a quiet act of resistance against the commodification of creativity. But it was still a centralized choice: I controlled the contract. Manus faces the same dilemma. It can choose to be a good actor, but it cannot choose to be trustless.
Contrarian: The Counterintuitive Case for the Block
One might argue that the regulatory block is actually a good thing for decentralization. After all, it prevented Meta from locking up the talent and technology. If Manus had been acquired, the agent would have become a closed product. Now, it remains independent. Tencent, despite its size, is not yet a monopolist in the AI agent space. The Singapore base gives Manus access to a global market without the baggage of a Chinese state-owned identity.
But here is the contrarian angle: the block itself is a form of centralization. It reinforces the power of the state to decide who can own what. In a truly decentralized world, no single entity would have the authority to block an acquisition. The network would decide. The users would vote with their feet. If Manus were a DAO, the Meta acquisition would have been put to a vote. The community would have weighed the pros and cons. Instead, a handful of officials made the call.
Furthermore, the block may actually accelerate the consolidation of AI agents under state-aligned capital. Tencent now has a foothold in a leading agent platform. Soon, it may push for integration with WeChat, Tencent Cloud, and enterprise tools. The agent will become a funnel for the Tencent ecosystem. That is not independence; it is a different kind of dependency.
The real lesson from Manus is that we need to build AI agents that are not subject to any single jurisdiction or corporate parent. We need agents that run on decentralized infrastructure, where the governance is transparent, the data is encrypted and user-owned, and the decision-making is on-chain. This is not a pipe dream. Projects like Autonolas, Fetch.ai, and others are exploring exactly this. But they are still early. The Manus case shows that the market is ready for a decentralized alternative.
Takeaway: Follow the Fear, Not the Chart
Follow the fear, not the chart. The fear here is that AI agents will become the new battleground for corporate and state control. The chart shows a bull market for AI, but the underlying architecture is still centralized. If you are building or investing in AI agents, ask yourself: who has the power to shut this down? Who can change the rules? If the answer is a person or a board, then you are building on sand.
If you can design a system where the agent’s logic is executed by a smart contract, where its data is stored on a public blockchain, and where its governance is controlled by a token-holding community, then you have built something that no regulator or corporation can take away. That is the sovereignty of code.
Manus’s story is not over. It will go to Singapore, build its product, and likely succeed. But it will always be at risk. The next buyer, the next regulator, the next geopolitical shift could change its fate. The only way to achieve true autonomy is to eliminate the need for permission. That is the promise of blockchain. And that is the lesson we must carry forward.
I have seen the collapse of Terra-Luna, the crash of 2022, and the slow recovery since. In each case, the projects that survived were those that had a resilient community and a transparent governance model. Manus, despite its talent, lacks that. It is a reminder that in a world of centralized power, the only safe harbor is a protocol that no one controls.