TaiwanArtificial Intelligence Basic Act (人工智慧基本法)
In force.
An AI governance rule binding public and private bodies.
- Criminal exposure
- No
- Private right of action
- No
- Obligation class
- Governance
- Audit expectation
- none
What it requires
- An AI product or system that a central competent authority for the relevant industry, in consultation with the Ministry of Digital Affairs, identifies as a high-risk application must come with advisory notices or warnings (Article 5).
- Apart from that warning, this Act does not itself impose an operative disclosure, labeling, or risk-classification duty on an AI developer or deployer; it directs the government to bring existing law into conformity with, or adopt new rules under, its principles within two years of this Act's effective date.
- Where a sector regulator adopts a binding rule under this Act's risk-taxonomy mandate, follow that regulator's own implementing regulation for the specific duty it imposes rather than this Act's own text.
These lines are LexLint's reading of the law's text.
In the frameworks
- NIST AI RMF
- MEASURE 2.8Risks associated with transparency and accountability – as identified in the MAP function – are examined and documented.
Our reading of where this law's duties sit in each framework. It never says that running a control meets the law. How we read a framework against the law
Who enforces it
Enforcement body
National Science and Technology Council (central level), with special municipality, county, or city governments at the local level
What it makes you log
Logging duty
The Act directs the government to develop an AI risk taxonomy and risk-based management regulations (Article 16) and to bring existing law into conformity with its principles within two years (Article 18); read in full, it states no duty for a developer or deployer to log, record, retain, or later produce evidence of AI system operation. Article 4(5) names Transparency and Explainability as a governing principle for the government to apply, calling for AI outputs to carry appropriate disclosures or labels, but this is a principle for future rule-making rather than an operative record-keeping duty in the Act's own text.
- Kind
- None
- As of
What this law does
This 20-article framework act sets the government's guiding principles for AI development and application (Article 4: sustainable development and well-being, human autonomy, privacy protection and data governance, cybersecurity and safety, transparency and explainability, fairness and non-discrimination, and accountability), designates the National Science and Technology Council as the central AI competent authority with local governments at the municipal or county level (Article 2), and directs the Ministry of Digital Affairs to build a risk taxonomy and assessment framework that sector regulators use to write risk-based management regulations (Article 16).
For high-risk AI application, Article 17 directs the government to clarify liability attribution and establish relief, compensation, or insurance mechanisms, rather than creating those mechanisms itself. Article 5 provides that an AI product or system identified as a high-risk application by a central competent authority for the relevant industry, in consultation with the Ministry of Digital Affairs, shall come with advisory notices or warnings.
Article 18 gives the government two years from this Act's effective date to bring existing law into conformity with it or enact rules it requires, so a crawler or training-data, agent-transaction, or automated-decision duty for a specific AI activity, and any disclosure or labeling duty beyond the Article 5 warning, is not yet a duty this Act itself imposes. The Act provides that it takes effect from the date of its promulgation. It was promulgated on .
Under Article 13 of the Central Regulation Standard Act, a law that takes effect from its promulgation takes effect from the third day counting from the day of promulgation.
When LexLint raises it
When your app profile says your app generates content with AI, deploys a chatbot or makes high-risk automated decisions.
Back to the example · Lint your app
Page updated .