AI Regulation Tracker  /  Rule pending

Taiwan's AI Basic Act signals a coming 'I am AI' labeling duty, with the sector standard still pending

Taiwan's new framework statute signals that firms deploying generative AI or AI chat will have to disclose AI-generated content and AI identity to consumers. The exact labeling standard and start date are still pending from regulators.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

Taiwan has put a national duty framework on the books, and one of its quieter provisions may reshape how businesses present anything an AI produces. The Artificial Intelligence Basic Act (人工智慧基本法) cleared its third reading in the Legislative Yuan on December 23, 2025 and was promulgated on January 14, 2026. It is a framework statute, a 基本法, which means it sets principles and points each ministry toward writing the detailed rules. Among those principles is a transparency standard that lawyers reading the text expect to harden into a concrete "I am AI" labeling duty.

What the statute actually says

The Act's transparency principle states that AI outputs should provide appropriate information disclosure or labeling so that affected people can assess risk and understand potential impacts on their rights. A separate provision addresses high-risk AI: after a product or system is designated high-risk by the relevant central sector agency in consultation with the Ministry of Digital Affairs (moda), it should explicitly mark precautions or warnings. A third provision directs sector agencies to build risk-based management rules and to help their industries develop guidelines. Read together, these give regulators a clear mandate to require conspicuous AI labeling, without yet specifying the format.

The emerging duty, as the firms read it

Law-firm alerts summarizing the Act describe an expected duty with two faces. First, businesses that generate content with AI, whether images, video, audio, or text, would be expected to attach a visible warning or disclosure label (a 警語 or 揭露標示) so consumers know the material is machine-made. Second, businesses that use AI to interact with customers, such as AI customer service, would be expected to disclose that the counterpart is an AI rather than a human, an "I am AI" signal that protects the consumer's right to know. For high-risk applications, the reads go further, pointing to warning labels plus prior review or third-party verification before deployment.

What it does not do yet

This is the part professionals should hold onto. The Basic Act carries no direct penalties of its own; enforcement runs through each ministry's sector rules. More importantly, the exact labeling technical standard and its effective date have not been issued. The reporting here attributes the specific "must label" and "must disclose identity" duties to law-firm interpretations of the Act, not to a published sub-regulation. moda and the sector regulators still have to operationalize the standard, so the shape of a compliant label, the placement, the wording, and the start date remain open. Building a rigid compliance product against a standard that does not exist yet would be premature.

Why a US professional should track this

For US firms operating in Taiwan, the direction is unmistakable even if the detail is not. Taiwan is choosing a principles-first, enabling statute that defers hard duties to sector regulators, a contrast with the European Union's binding risk tiers and fines. But on transparency it converges with a global pattern. The EU AI Act's Article 50 requires disclosure of AI-generated or manipulated content and of AI systems that interact with people. China's deep-synthesis rules mandate labeling of synthetically generated media. California has enacted AI transparency measures in the same vein. Taiwan's labeling regime, once its sub-rules arrive, becomes the next data point in an Asia-Pacific move toward mandatory AI disclosure, and a preview of the compliance posture multinationals will need across markets.

What to do in the meantime

The practical work is preparation, not implementation. Map where your organization uses generative AI or AI chat with the public, and treat customer-facing chat and published AI media as the two areas most likely to draw an early labeling rule. Note which outputs are fully machine-generated and which are human-reviewed, since a human-in-the-loop record may shape how a disclosure has to read. Keep a channel open to moda announcements and to your sector's competent authority, because the operative duty will arrive through their rules rather than through the Basic Act itself. Review any vendor contracts for AI customer-service or content tools so responsibility for a compliant label is assigned rather than assumed. When the technical standard publishes, you will be positioned to label quickly rather than starting from a blank inventory, and you will avoid retrofitting disclosures across a system you never fully mapped.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.