Taiwan's AI Basic Act: a promotion-first framework law takes effect

Asia gained its second comprehensive AI statute within three weeks of its first. Eight days before South Korea's binding AI Basic Act took effect on 22 January 2026, Taiwan's own AI Basic Act entered into force on 14 January 2026. The two share a name and a region, but not a method: where Seoul wrote enforceable duties for operators, Taipei wrote a framework that binds the government, not the market.

How it became law

Taiwan's Executive Yuan (the Cabinet) approved the bill on 28 August 2025. The Legislative Yuan passed it in its third and final reading on 23 December 2025, and President Lai Ching-te promulgated it on 14 January 2026. Under Article 20, the Act takes effect on the date of its promulgation — so passage and entry into force are, for practical purposes, the same 2026 event. The statute is short by design: 20 articles that set direction rather than impose a compliance regime.

A framework, not a rulebook

This is the decisive feature for anyone comparing it to the EU model. The AI Basic Act is a basic law (基本法) — a constitutional-style instrument that states principles and obligations for the state, and instructs ministries to build the operational rules later. It does not create high-risk categories, conformity assessments or fines for private operators the way the EU AI Act does. Instead it commits the government to promote AI while protecting rights, and delegates the detailed machinery to the competent authorities.

The National Science and Technology Council (NSTC) is the central competent authority (Article 2) — a contested choice, since earlier drafts left the authority unnamed and the Ministry of Digital Affairs (MODA) was the other candidate. In the final settlement NSTC holds the centre while MODA carries key technical mandates, most notably the risk-classification work.

Seven principles

The Act codifies seven governance principles that recur across international AI instruments: sustainable development and well-being; human autonomy; privacy protection and data governance; cybersecurity and safety; transparency and explainability; fairness and non-discrimination; and accountability. These are the same family of principles found in the OECD AI Principles and the UNESCO Recommendation — Taiwan is grafting its statute onto the shared global grammar rather than inventing a new one.

What the government must build

Several articles convert the principles into assignments for the state:

• Risk classification (Article 16). MODA must develop an AI risk taxonomy

and assessment framework aligned with international standards, and competent authorities are to help their industries draw up sector guidelines. This is where Taiwan's eventual, EU-style risk tiering will live — but it does not exist yet; the framework has to be written.

• National AI strategy (Article 6). A National AI Strategic Committee,

convened by the Premier and drawing in scholars, industry and officials, must meet at least annually to review the National AI Development Guidelines.

• Innovation environments (Article 11). The government may establish safe

testing environments — regulatory sandboxes — for AI research and services.

• Data governance (Articles 13–14). Mechanisms for open data, data sharing

and reuse, paired with personal-data protection during AI development.

• Labour (Article 15). Safeguarding workers' rights, bridging skills gaps

and offering employment counselling to those displaced by automation — a social dimension more prominent here than in most AI statutes.

• Liability (Article 17). Clarifying attribution of liability and providing

relief, compensation or insurance mechanisms for high-risk applications.

Where it sits in the global map

Taiwan deepens the pattern traced in our analysis of international AI governance beyond the EU and of the two Asia-Pacific models. South Korea legislated binding operator duties; Singapore relies on voluntary frameworks; Japan chose a promotion-first law resting on guidance. Taiwan lands closest to Japan: a light, innovation-oriented statute that sets values and institutional scaffolding and leaves the binding detail to later instruments. For a European organisation operating in Taiwan, the near-term consequence is modest — there are no new direct obligations on day one — but the direction is set, and the risk-classification framework that MODA now has to build is the part worth watching. As in Korea, the lesson is to build governance on the shared risk-based grammar, so that each jurisdiction's later overlays become adaptations rather than rebuilds.