Thank you, everyone, and thank you to Audrey for being here virtually. We are very happy to have this special dialogue with Audrey Tang.
As you all know, Audrey is famous for making contributions not only to AI but also to web3 and many kinds of technology, not only in Taiwan but around the world.
Allow me to introduce myself. My name is Ju Chun Ko. I am a legislator in Taiwan, and I am really into crypto and the future.
Today we want to talk not only about web3 and cryptocurrency but also about AI. Audrey is now introducing Civic AI: 6-Pack of Care around the world, perhaps in the future not only on Earth but on the Moon or Mars.
We want to talk about legislation and regulation for AI and cryptocurrency. In the past year, Taiwan passed the AI Basic Act and the Virtual Asset Service Act. It is a small step in Taiwan’s Parliament, but a giant leap for Taiwan and for Asia.
Audrey, would you like to say hello to the audience?
Good local time, and thank you, Ko, and everyone here. I am happy to join you again, this time virtually from Taipei.
This is not our first conversation. It feels like catching up because I have been on Ko’s podcast multiple times.
This milestone belongs to you and your colleagues across every caucus in the Legislature, and to the Cabinet in Taiwan, whose draft was merged alongside yours in the Finance Committee. I am joining today as a fellow practitioner. I did not write the Act with you, but I celebrate this joint milestone.
My first question is about your feeling on the recent movement in Taiwan.
Starting from 2008, we had Bitcoin and blockchain. Around 2013, there were hackers and developers in Taiwan and Japan, but at that time there was almost no regulation. By 2016, people could buy Bitcoin in convenience stores in Taiwan. At that time, we were a kind of leader.
In the past, Taiwan’s government was a little conservative. Later, we saw many countries establish their own crypto regulations and legislation about blockchain, while Taiwan remained in a wait-and-see position.
But in the first half of this year, we passed Taiwan’s first Virtual Asset Service Act. It is a kind of clarity moment for Taiwan. Earlier, in January, President Lai Ching-te announced Taiwan’s first AI Basic Act.
What is your feeling about that?
Honestly, as I said, it is about celebration.
For years in Taiwan, we had this old conversation. We said that something had “not yet reached social consensus.” But that phrase was never neutral. This grey area quietly handed power to whoever could already operate comfortably in that grey zone, because delay always favours incumbents.
You mentioned, for example, buying Bitcoin in convenience stores in 2016. That era showed that Taiwanese people were ready long before the rules were ready.
The shift you have led is not from conservative to permissive. It is from ambiguity to clarity. Ambiguity is a subsidy to whoever already holds interpretive power. The AI Basic Act and the new Virtual Asset Service Act take that subsidy away.
You often say that society should shift from regulation to cybernetics, toward development and innovation, in cryptocurrency and blockchain.
Taiwan is a late follower in regulation. The European Union had its AI Act long before Taiwan’s AI Basic Act. Taiwan produces semiconductor chips for the world and helps generate the future of AI, but we only passed our first AI Basic Act this year.
For virtual assets, we had our first clarity moment last month. What is the advantage of being a late follower in regulation? Now that we have regulation, how do we make innovation happen in Taiwan and connect Taiwan with Japan and other places? What keeps regulation from becoming a restriction on innovation?
That is a great question.
As Cyber Ambassador, I often return to the origin of “cybernetics”. It comes from the Greek word for steering. It is more important to hold the steering wheel with the people, allowing rapid pivots when rules no longer match the underlying technology. The speed of steering is more important to me than the speed of acceleration.
What we are discussing in AI is exactly like that. The test of good regulation is not whether today’s AI settings are perfect, because they never are. We are already seeing small-scale recursive self-improvement. A larger model such as GPT Sol can train a smaller model such as Luna. Imagine if the direction were reversed.
We should implement the AI Basic Act so that the risk profile remains adaptable. Instead of mandating levels one through five, like a fixed autonomous-driving ladder for AI agents, we should allow different classes of integration. Japan also takes this kind of approach, with different adoption types in different communities.
The point is to put AI in the loop of humans, not humans in the loop of AI. Each sector and each community should determine how to integrate AI into its loop, instead of being accelerated out of its capacity to pivot.
Taiwan’s approach, like Japan’s, should not be to retrofit rules after a crisis. We can learn from other jurisdictions, including the European Union, while preserving the capacity to steer instead of being fixed to any one setting.
When we talk about technology or technology policy, I have a deep belief that it has no border and no colour. It is transparent, bipartisan and tripartisan. Different parties in Taiwan and around the world should support technology policy that helps people live better with technology.
You serve as Cyber Ambassador under the current government. I serve as an at-large legislator for an opposition party. We are both working hard on technology policy, including the AI Basic Act and the Virtual Asset Service Act.
We are forming a consensus about technology policy in Taiwan, but you are trying to do something bigger. You are introducing Civic AI and the 6-Pack of Care around the world, perhaps one day even beyond Earth. Can you introduce that vision to our audience?
Yes, thank you. The framework is available at civic.ai, with very nice Japanese illustrations.
Our Taiwan Model-inspired Civic AI starts with particular communities. It starts with the people we care about. It does not begin in interstellar space. It begins in particular communities here on Earth.
AI systems can be extractive. Data becomes like oil, drilled by companies from communities into one giant cloud singleton model. That model sees everything and trains on the data embedded in communities, but the data is lifted out and loses its context.
Civic AI goes the other way. We say data is soil, tilled by each community that can have its own local Kami, or knowledge artefact management intelligence, combining knowledge management and AI. Imagine 8 million Kamis, one in every community.
In fact, your OpenClaw instance published the readings and transcripts of your work on the Virtual Asset Service Act. My Kami read them days ago and briefed me on the improvements. That is a good way to bridge partisan and cultural differences.
Although I am nonpartisan and do not belong to a political party, people in Japan are also looking at this kind of cross-partisan use of AI, as we have seen in the Diet.
We are at WebX, not only web3 or web4 but WebX, at the frontier of a new kind of internet.
We might imagine connections among AI, the new internet, web3, blockchain, virtual assets and cryptocurrency. Japan’s ruling Liberal Democratic Party recently released an AI white paper that mentioned agentic commerce.
Tell us more about your Civic AI, your Kami and your AI agent. Have you talked with your Kami about agentic commerce? Is there something interesting that connects AI agents with virtual assets or blockchain?
I would ask AI systems the same four questions we have been asking web3 services:
Can people move between providers? Can people inspect the rules, including smart contracts? Can communities govern their own deployment? Is somebody answerable when the deployment breaks?
An agent buying and selling on my behalf must pass all four tests. I need to be able to see the rules it follows.
With AI systems and agentic coding, there is a danger. An agent can hand you 70,000 or 70 million lines of code and say, “That is exactly the rule I follow.” Nobody has enough time in one lifetime to understand all 70 million lines.
What do we do? I have been working with LemmaScript, an annotated version of TypeScript. It can take existing code, such as the roughly 70,000 lines of JavaScript in EtherCalc and SocialCalc, which I maintain with spreadsheet inventor Dan Bricklin, and let us write invariants as cryptographically attested lemmas that AI agents can prove.
There are specialised systems such as Leanstral that can inspect those lemmas, prove them and produce new verified environments automatically.
For the first time, the same hardening we have applied to smart contracts and on-chain protocols can be used to triage and harden the output of the same agentic systems that perform vibe coding.
Once each layer is hardened, we do not need to inspect every line. We can run the code through Dafny or other proof assistants and prove that the long program satisfies a small, manageable set of invariants.
I think this is a very exciting application at the intersection of cryptography and AI.
That is frontier work led by you, and we look forward to seeing more.
The word “frontier” also matters to an economy or a country. We think about frontier AI, and frontier AI is now part of government-to-government relationships. A few weeks ago, one frontier model called Fable was forbidden for some other users from using it.
When we think about a frontier, we think about opportunity and risk for an economy or a country. You are a thought leader on frontier ideas. What frontier or mainstream models does your on-chain Kami use every day?
How might frontier models combine with a frontier blockchain such as Ethereum? There may be government officials here, or they may later watch the recording. What should they think about when AI and blockchain are combined?
It is a great question. First and foremost, the frontier question is about resilience.
My Kami is deliberately model-agnostic. It runs on a mixture of models. In Japan, a team called Sakana AI developed the same idea. They call it Fugu, and it is now offered as a service on OpenRouter. There are also open-source implementations such as OpenFugu.
The idea is to train a very small model, perhaps 7 or 8 billion parameters, on your actual usage so that it can run at the edge. It learns how you use models and assigns the best verifier, planner and executor for each turn.
Every time I ask the Kami to do something, it chooses three different models for those three roles within the turn.
The advantage is similar to Ethereum having many independent implementations. Removing one implementation does not crash the chain. Removing one model does not crash the Kami.
This is where the on-chain layer can contribute to the frontier conversation. Even when the underlying model changes, the agent’s identity, reputation, record and rules live on an inspectable substrate. The record cannot simply be changed once it is there. You do not lose accountability by putting the agentic record on-chain.
If the record exists only with a frontier-model provider and that provider removes the model, the whole world may lose contact with the agentic surface.
My message to policymakers is not to choose one frontier model and chain yourself to it. Test whether citizens and agencies can inspect the rules, switch providers, govern their own layer and hold somebody answerable whenever any part breaks.
Frontiers come and go. What compounds is the data soil and the steering capacity we keep.
I think that is a great idea.
I was the lead sponsor of Taiwan’s AI Basic Act and a principal co-sponsor of the Virtual Asset Service Act, so I would like to go further.
You appear on a screen here in Tokyo and look like Audrey, but there may be no way, or perhaps a new way, to prove that you are the real Audrey. Perhaps your Kami could pretend to be Audrey in real time.
What should identity look like in the AI and blockchain era? Years ago, you proposed decentralised identity in Taiwan and initiated a digital-wallet project. In 2026, companies such as World are introducing proof of humanhood for the AI era, and even an agent book where a Kami or AI agent can appear online and on-chain.
We need not only decentralised identity for human beings but also identity for an AI agent, proving that it is authorised by Audrey. What comes next for identity and proof of agency?
Proving that I am not a deepfake is actually the easier part. Humans already have strong solutions, including local biometrics and signatures.
The next move is dual. It is not only proof of personhood for humans but proof of delegation and authorisation for agents.
On-chain, you can link
audreyt.ethtojdd-kami.eththrough a soulbound token (SBT). It gives me a soulbound reputation path to JDD Kami. That mechanism is already there.The larger question is how to make that accountability real rather than symbolic.
The design principle I would offer is that identity is not only about proving what you are. It is about proving who answers for you.
I would let communities, rather than any single provider, including World.org, govern that layer. One useful clarification is that World.org is both an implementation of an underlying protocol and a reference implementation of an open protocol that other people can use.
I would refrain from saying that one front end or one particular method of interaction should be the only way. Once it becomes the only way, the same vendor lock-in problem enters the picture.
We should apply the same four tests. We should not let ambiguity subsidise whoever already holds interpretive power. We should make data soil, and make freedom to migrate and move between services the default in our daily use, in our Kamis and in our regulations.
You wrote your legislative notes as a letter to the Taiwan of 2036. The test is whether the people who inherit a system ten years from now can still fix it when it breaks.
That is the original software freedom: if it breaks, you keep both parts, and you can put them together in a different way.
Thank you for reading my blog. If you are interested, you can find it online at juchunko.com. There is an article about Taiwan’s first Virtual Asset Service Act.
Today our topic is the state and blockchain. We have had a conversation between you and me, and between me and your Kami.
We are coming to the end of this special dialogue. Technology, AI and blockchain carry risk, but if we contribute enough, we believe technology can bring opportunity and prosperity to a country, an economy and individuals.
You mentioned my thinking about 2036. What is your thinking about the next chapter for a state or economy, and for individuals, as blockchain enters the AI era?
The next chapter is being written together by the people in this room, more than by any official or ambassador.
The laws we have seen around the world, including those you and your colleagues passed this year, are not the finishing line. I was reading the risk-profile taxonomy published less than a week ago by the Ministry of Digital Affairs for implementation of the AI Basic Act. It is really an invitation.
What happens next depends on whether actual builders treat this clarity as soil to grow in rather than a ceiling to work around.
My hope for the next chapter is simple. We keep AI in the loop of communities. We keep money and identity on substrates that people can inspect and leave. We keep every system fixable by the people who inherit it.
If we hold to that, the risks you named become manageable and bounded, while the prosperity you named becomes shared and unbounded, for Taiwan, for Japan and for everyone building between them.
Thank you. It is always a pleasure to have a conversation with you, and congratulations once more to you and your colleagues in the Legislature across every caucus.
Audrey joined us virtually today. She will have another session this afternoon, and I will also have another session at the same time on a different stage. Please join Audrey’s session and mine this afternoon.
Thank you all for joining this morning’s special dialogue. I hope all will be well with you, both in the real world and the virtual world. Thank you very much. Until we meet again.
Thank you.