Yat Siu Warns Crypto AI Agent Users Could Lose Their Data Overnight

Yat Siu didn’t mince words. The Animoca Brands chairman told the Token2049 Singapore crowd that anyone leaning on an AI agent for crypto trading could wake up one morning and find the whole thing gone — no warning, no appeal, no backup.

The risk isn’t the algorithm. It’s the company behind it. Siu’s point was pretty simple: if the service provider suspends your account, your agent disappears with it. And so does everything that agent learned about you — your preferred markets, your risk tolerance, the months of back-and-forth that shaped how it traded on your behalf. For a casual user, that’s annoying. For a trader who’s spent a year building a personalized system, it’s a real problem. Siu said users should have the right to download their data and move it somewhere else so another agent can pick up where the last one left off. That’s not a feature most platforms currently offer.

When Agents Get Too Narrow

Siu knows the shortcomings firsthand. His own agent became so laser-focused on crypto trading and investment management that it started reading unrelated questions as crypto-related. Ask it about dinner plans, get a portfolio suggestion. He said running several specialized agents — each with a defined job — can help avoid that kind of drift.

One agent can become too narrow. That’s basically the whole problem in six words.

He also warned against piling too much onto a single assistant. Personal scheduling, family plans, travel logistics, crypto advice — dump all of that into one system and you’re not just confusing the agent, you’re probably confusing yourself about what it’s actually doing. A 2025 CoinGecko survey found 14.5% of respondents would hand over full portfolio management to an AI agent. Siu’s caution seems aimed squarely at that group.

And there’s a subtler issue with delegating analysis. If a thousand traders ask the same AI the same question about the same trade, they’ll probably get roughly the same answer. The edge disappears. Siu said personal follow-up — bringing your own knowledge, your own concerns, your own context — is what produces different responses. That’s where the personalization actually lives, not in the initial prompt.

Coinbase, Binance and the Race to Give Agents Real Tools

The industry isn’t waiting around. Coinbase rolled out Agentic Wallets in February, letting agents spend and trade crypto within limits users set themselves. Binance followed in March with AI Agent Skills, giving agents live market data and the ability to place and manage spot orders. Both moves push agents from passive monitors into active participants — which raises the stakes considerably if access gets cut.

Read also: Bitcoin’s $82,000 Weekly Close Puts $80,000 at Risk as Coinbase Demand Hits June Low

Siu drew a distinction worth keeping in mind. An agent isn’t just a chatbot. A chatbot answers when you ask. An agent watches, notices something relevant, and flags it before you even think to ask. He used conference planning as an example — an agent spots a newly announced event and recommends you go, without waiting to be prompted. That proactive behavior is useful. It’s also what makes losing the agent so disruptive.

But useful doesn’t mean safe. Siu urged people to think hard about what they’re actually feeding these systems. He compared an AI assistant to a third-party adviser and said users should ask whether the agent really needs to know a particular detail to answer a question. Finances, income, relationships — an agent that gets all of that builds a surprisingly complete picture of its user. Most people probably don’t think about that trade-off when they’re typing.

Regulators Are Watching the Clustering Problem

It’s not just individual traders who should worry. In April 2025, the Bank of England’s Financial Policy Committee flagged that widespread use of similar AI models could push investors toward related positions during stress periods — and amplify market shocks in the process. Similar prompts, similar answers, similar trades. The systemic angle is still murky, but the concern is real.

Siu’s view is that personalization is the partial answer. An agent that learns a user’s preferences over time can start trading in line with those preferences — building something genuinely individual rather than a generic strategy that thousands of other users are running in parallel. That stored history matters. The agent remembers past conversations, doesn’t need the same instructions twice, and can surface information without being asked.

More context: Hayes Says AI Bailout Risk Could Drive Bitcoin Higher as ETF Outflows Hit $228M

Losing that history is the thing Siu kept coming back to. He described the possibility of an assistant disappearing after years of accumulated knowledge. The information stored might be impossible to reconstruct. And right now, there’s no standard mechanism to prevent it — users can’t just export their agent’s memory and plug it into a competitor’s platform.

A 2026 Pew Research Center survey found 71% of U.S. adults believe increased AI use will make their personal information less secure. Siu’s argument is that crypto traders face a version of that problem that’s both more immediate and more financially consequential than most.