Behind the Scenes of Taiwan's New Crypto Law: A Dialogue between Audrey Tang and Ju-Chun Ko at WebX2026

On July 13, 2026, a special dialogue took place on the CRYL stage of WebX 2026 between Ju-Chun Ko, a member of Taiwan's Legislative Yuan, and Audrey Tang, Taiwan's first Digital Minister and current Cyber Ambassador. Tang participated remotely and discussed the regulatory philosophy surrounding the Virtual Asset Service Provider Act (VASP Act) and the AI Basic Act, both established in the first half of 2026, as well as the concept of "Civic AI" and the integration of AI agents with blockchain.

Transition from Ambiguity to Clarity

Despite the environment for purchasing Bitcoin at convenience stores in Taiwan being established in 2016, regulatory development lagged for the next decade. However, in 2026, the VASP Act (established in June) and the AI Basic Act (enforced in January) were successively established, significantly changing the situation. When Ko inquired about the significance of this turning point, Tang pointed out the structural issues of the "gray zone."
Audrey Tang
The expression "not reaching social consensus" may seem neutral at first glance, but it is not neutral at all. The gray zone has quietly continued to empower those who can already act within it. Delays always benefit vested interests.

This transition is not a shift from "conservatism to liberalization"; it is a transition from "ambiguity to clarity." Both the AI Basic Act and the VASP Act remove the subsidies for interpretation rights that have been monopolized by a select few.
Note: Taiwan's VASP Act was established on June 13, 2026, and stipulates a registration system, customer protection, and anti-money laundering measures for virtual asset operators. The AI Basic Act (enforced on January 14, 2026) is a principle law consisting of 20 articles, clearly stating seven principles: sustainability, human autonomy, privacy, cybersecurity, transparency, fairness, and accountability. Neither law contains direct punitive provisions and aims to balance innovation promotion with rights protection.

Advantages of Latecomer Countries and the Philosophy of Regulatory Design

While leading the world in semiconductor manufacturing, Taiwan has been a latecomer in AI regulation compared to the EU. When Ko asked about the "advantages of latecomers" and the keys to ensuring that regulation does not stifle innovation, Tang discussed the origins of the term "cybernetics."
Audrey Tang
Cybernetics comes from the Greek word "kybernetes" (helmsman). The speed of steering is more important than the speed of acceleration.

The standard for good regulation is not whether today's AI settings are perfect. We are already entering a stage where large models can train small models. It is essential that risk profiles are adaptable rather than fixed level classifications.

Like Japan's approach, we should establish different integration types for each community, designing it so that AI enters the human loop rather than humans entering the AI loop. Each community can incorporate AI at their own pace. Both Taiwan and Japan must maintain steering capabilities rather than retrofitting regulations after a crisis occurs.
Note: The Japanese government's "Artificial Intelligence Basic Plan" (decided by cabinet in December 2025) aims to make Japan "the easiest country in the world to develop and utilize AI," with a second phase revision plan decided in July 2026. The Liberal Democratic Party's AI White Paper 2.0 (April 2026) positions "Japan AX (AI Transformation)" based on agent AI as a national goal.

The Design Philosophy of Civic AI: "Data is Soil"

When Ko prompted Tang to introduce her "Civic AI" framework (civic.ai), Tang explained the fundamental difference from general AI development models in terms of "data handling."
Audrey Tang
AI systems can draw data from communities like oil and pour it into a massive cloud model, losing context along the way. Civic AI is the opposite idea. "Data is soil and never oil."

Each community has its own small knowledge management and AI, referred to as "commi." In fact, my commi recently absorbed the results related to Ko's virtual asset law. Bridging across parties and cultures can be realized.

Identity Verification in the Age of AI

The discussion evolved into decentralized identity (DID) and agent verification, stemming from the question, "Is the person on the screen really Audrey Tang, or could it be a real-time deepfake?"
Ju-Chun Ko
Companies like World are providing "Proof of Human" and introducing systems like "Agent Book" that allow AI agents to be registered on behalf of humans. We will need a mechanism not only to prove the identity of the human but also to certify the AI agents you have authenticated.
Audrey Tang
Human identity verification is almost resolved through a combination of local biometric authentication and signature technology. The challenge lies in the next stage. We need not only "Proof of Personhood" for humans but also "Proof of Agent" for AI agents. If my commi transacts on my behalf, how do we prove that the agent is genuinely authenticated by me?

"Hardening" AI Agents with Blockchain

Tang revealed that her commi "JTEAMI" is already operational on Ethereum and discussed an approach to make AI agents verifiable using methods similar to smart contracts.
Audrey Tang
My commi JTEAMI operates on a permissionless agent protocol as ERC-8004, with Agent ID 22714. If an AI agent trades on my behalf, it must pass four tests: portability among providers, verifiability of routes including smart contracts, governance of deployment by the community, and accountability in case of failures.

The problem is that no one can read 70 million lines of code. This is where formal verification comes into play. Using LinuxScript, we can write out existing code's "invariants" as cryptographically provable lemmas. Proof assistants like Lean and Dafny can prove these and automatically generate new invariants. The same robust methods proven in smart contracts can be applied to AI agents.

Additionally, my commi is designed to be model-independent, mixing multiple models. Just as Ethereum has multiple independent implementations, if one model fails, the whole system does not stop. Frontier models come and go, but the soil of data and steering capabilities accumulate. What I want to convey to policymakers is not to be bound by a single frontier model but to ensure that citizens and institutions can inspect their own rules and maintain accountability for everyone.
Note: ERC-8004 is a proposed standard for managing AI agent identities on Ethereum. The Liberal Democratic Party's AI White Paper 2.0 (April 2026) positions "agent AI" as a key technology to change national structures, and Tang's remarks connect with policy discussions in both Japan and Taiwan.

Session Summary

As a closing remark, Ko expressed the belief that technology policy should transcend borders and party lines. He emphasized the significance of jointly advancing the same laws beyond the positions of the ruling and opposition parties.

"Tech policy must be transparent and bipartisan. As an opposition member, I, along with you as the current government's Cyber Ambassador, have advanced Taiwan's AI Basic Act and VASP Act together. That is the message I want to convey today."

As a country that has contributed to the world in semiconductor manufacturing while being a "latecomer in regulation," Taiwan's establishment of a clear legal framework for both AI and cryptocurrency serves as a highly valuable reference for policymakers in Asia, including Japan.