Taiwan’s Artificial Intelligence Basic Act (AI Basic Act) was officially promulgated in January 2026 (see 3.1 General Approach to AI-Specific Legislation).

Regarding liability arising from the use of AI technology, please see 10.1 General Theories of Liability.

In recent years, the Executive Yuan of Taiwan has announced the Digital Nation and Innovative Economic Development Plan and the Taiwan AI Action Plan – currently, the Taiwan Artificial Intelligence Action Plan 2.0 (2023-2026) – to express the Taiwan government’s desire to develop world-leading AI infrastructure for device solutions and to establish a sound ecosystem that creates a niche market.

As Taiwan is well known for information and communications technology, and the semiconductor industry has established a good foundation for intelligent technology development, the government has been encouraging companies to become essential partners in the value chain of global AI technology and intelligence systems. It will leverage the advantages in hardware and software techniques to promote AI technology, whether it be generative or predictive AI, across industries with test fields, regulations, data-sharing environments, etc.

In August 2022, the Ministry of Digital Affairs (MODA) was formally established to handle matters relating to facilitating Taiwan’s digital development of its telecommunications, information, cybersecurity, internet and communications industries, co-ordinating national digital policies, supervising national cybersecurity policies, managing communications and digital resources, and assisting digital transformation. According to the Organization Act of the Administration for Digital Industries, MODA is in charge of providing guidance and incentives for the interdisciplinary digital innovation of AI, big data, platform economy and other digital economy-related industries.

MODA’s Administration for Digital Industries has established the “AI Product and System Evaluation System” and “AI Product and System Evaluation Guidelines” to regulate AI products and systems and provide domestic AI product evaluation services. The goal is to protect the rights of product users and promote the positive development of the AI industry. The AI Evaluation Center will focus on developing evaluation tools and systems to meet domestic testing needs, and will engage in cross-departmental co-operation to evaluate different types of AI systems, products and services. By establishing a trusted AI environment, Taiwan aims to align with international standards, promote compliant AI products in the global market, and accelerate expansion into the global markets. The AI Evaluation Center will continue to analyse industry needs and trends, in order to build a safer and more reliable AI evaluation environment.

The AI Basic Act was officially promulgated in January 2026. It does not impose obligations on the private sector, but it does highlight the following key principles.

• The government shall avoid AI applications that:
1. infringe upon the life, body, liberty or property of individuals;
2. disrupt social order, national security or the ecological environment; or
3. result in violations of laws and regulations regarding matters such as bias, discrimination, false advertising, misinformation or fabrication.
• The government shall adhere to the principle of the best interests of children and adolescents. Where an AI product or system is determined to be a high-risk application by the sector-specific competent authority in consultation with MODA, clear warnings or precautionary statements shall be explicitly indicated. MODA and other relevant authorities shall provide or recommend tools or methods for assessment and verification, to facilitate each sector-specific competent authority in handling the aforementioned matters.
• To promote innovation and the sustainable development of AI technology, each sector-specific competent authority may establish or enhance innovative experimental environments for the research, development and application of AI products or services.
• The government shall establish mechanisms for data openness, sharing and reuse, to enhance the availability of data for AI use, and shall regularly review and adjust the relevant laws and regulations.
• The sector-specific competent authority, in consultation with the competent authority for personal data protection, shall avoid unnecessary collection, processing or use of personal data during the research, development and application of AI, and shall promote the incorporation of personal data protection into default and design-related measures or mechanisms, to safeguard the rights and interests of data subjects.
• The government shall actively utilise AI to ensure the labour rights and interests of workers. The government shall proactively bridge the skills gap caused by the development of AI, enhance labour participation, safeguard economic security, and implement dignified labour.
• MODA shall refer to international standards or norms to promote an internationally interoperable risk classification framework for AI, and shall assist each sector-specific competent authority in formulating risk-based management regulations. In addition, sector-specific competent authorities shall formulate risk-based management regulations and assist relevant industries in formulating industry guidelines and codes of conduct, as necessary for risk management of AI applications and in accordance with the aforementioned risk classification framework.
• The government shall clearly define the allocation of responsibility and conditions for liability in the application of high-risk AI, and shall establish mechanisms for remedy, compensation or insurance.

In early 2018, Taiwan’s Congress promulgated the Financial Technology Development and Innovative Experimentation Act (the “FinTech Sandbox Act”) for the fintech regulatory sandbox to promote fintech services and companies. The Act enables fintech businesses to test their financial technologies in a controlled regulatory environment. Although the FinTech Sandbox Act is not specifically designed for AI, machine learning or big data, the promoters of novel AI-related financial business models or big data technology may prove their theories and applications within the scope of the Act and enjoy exemptions from relevant laws and regulations.

In the same vein as the FinTech Sandbox Act, the Unmanned Vehicle Technology Innovation and Experiment Act (the “Unmanned Vehicle Sandbox Act”) was enacted in late 2018, establishing a regulatory sandbox for autonomous/self-driving vehicles. The Act aims to provide a better environment for AI application testing and IoT technology in transportation. The term “vehicle” under the Unmanned Vehicle Sandbox Act includes not only cars but also aircraft and ships/boats.

The objective of these two regulatory sandbox laws is to enable relevant businesses to test their innovative ideas and technologies within a permitted safe harbour or sandbox scope. In general, prior to entering the sandbox, applicants need to obtain approvals from the competent authorities. Once the experiment is executed, the experimental activities may enjoy exemptions from certain licensing requirements and legal liabilities.

After the approved experiments are completed, the relevant competent authorities will examine the results (see 2.2 Involvement of Governments in AI Innovation). If positive, the competent authority will review and amend current rules in order to legalise the tested business models or activities. However, the sandbox applicants might still need to obtain licences or approvals from the relevant competent authorities before they can conduct the tested business models or activities outside the sandbox.

The Executive Yuan has announced multiple guidelines and plans for AI developments, such as:

• the Digital Nation and Innovative Economic Development Plan;
• the Taiwan AI Action Plan;
• the AI Technology R&D Guidelines; and
• the 5+2 Plan.

However, considering AI still has the nature of “technology” and could be applied in various industries, there are no regulatory agencies that play a leading role in the actual enforcement and monitoring of AI technologies. The enforcement and supervisory tasks fall under the jurisdictions of the relevant competent authorities, as follows:

• the Ministry of Economic Affairs (MOEA) is the competent authority for the Unmanned Vehicle Sandbox Act;
• the Financial Supervisory Commission (FSC) is the competent authority for the FinTech Sandbox Act;
• for AI-related healthcare matters, the Ministry of Health and Welfare is the authority responsible for supervising related products and industries; and
• the Food and Drug Administration (TFDA) is responsible for regulating the system for the safety and quality of food, drugs, medical devices and cosmetics – it grants product registration and clinical trial approvals, monitors manufacturing and importation, and conducts safety surveillance activities on health-related products.

The matter is not applicable in this jurisdiction.

The matter is not applicable in this jurisdiction.

According to the letter dated 16 June 2023 issued by the Intellectual Property Office (TIPO) of the MOEA, using copyrighted works to train AI models may infringe on the reproduction right of the copyright owner, unless it falls under the reasonable use exceptions specified in the Copyright Act. Whether content generated by AI is protected by the Copyright Act depends on the presence of human creative input, with content lacking human creative input not being protected by the Copyright Act. In addition, reproducing and representing a work using an AI model does not create new copyright.

The TIPO has noted that whether a specific act constitutes copyright infringement still needs to be determined by the judicial authorities based on the specific facts of the case. It is worth noting that the Copyright Bureau is planning to issue guidelines to further clarify the reasonable use of training data, whether AI-generated works are entitled to copyright protection, and other important issues, which should be closely monitored.

Relevant laws and regulations are expected to be promulgated following the enactment of the AI Basic Act. For example, under the AI Basic Act, MODA shall promote a risk classification framework for AI (please see 3.1 General Approach to AI-Specific Legislation). Currently, MODA plans to use checklists to help various ministries and agencies examine their AI application scenarios, and to identify, assess and respond to associated risks. MODA has already begun discussions with several regulators – including the Ministry of Labour, the FSC, the Ministry of Health and Welfare, and the Ministry of Education – with the aim of having these agencies form the first wave to use the checklist for assessing AI application risks.

There have been no court decisions specifically addressing IP issues in relation to generative AI. However, according to relevant court decisions of the Intellectual Property Court of Taiwan concerning the results of computer software calculations based on input parameters, it is difficult to consider such results as being subject to protection under the Copyright Act of Taiwan because they are obtained on the basis of mathematical calculations and are not created by any “human”.

Please refer to 2.2 Involvement of Governments in AI Innovation.

In light of the significant potential impact of AI on financial institutions, the FSC has established specific policies and guidelines concerning AI, unveiling the “Core Principles and Related Promotion Policies for the Use of Artificial Intelligence (AI) by the Financial Industry” in October 2023. This was followed by the “Guidelines for the Use of Artificial Intelligence (AI) in the Financial Industry” (“Financial Industry AI Guidelines”) on 20 June 2024. These guidelines are designed to guide financial institutions in the deployment, application and oversight of AI technologies.

The FSC has specified that the Financial Industry AI Guidelines are intended as administrative guidance. Financial industry associations that have crafted their own self-regulatory frameworks for AI usage are encouraged to incorporate the essential elements and measures from these Guidelines into their regulations. In the absence of such self-regulations, financial institutions are advised to follow these Guidelines for the implementation, utilisation and management of AI systems.

Given that the AI Basic Act (see 3.1 General Approach to AI-Specific Legislation) does not impose obligations on the private sector, no enforcement actions have been taken in this regard.

The Taiwan government is aware of and has published multiple guidelines and plans for AI developments in the past few years. In September 2019, the Ministry of Science and Technology under the Executive Yuan of Taiwan announced the AI Technology R&D Guidelines. As AI developments may bring revolutionary changes to various aspects of human existence, the Guidelines set forth the following three core values of which AI technology participants should always be aware when conducting relevant activities and endeavouring to build an AI-embedded society:

• human-centred values;
• sustainable developments; and
• diversity and inclusion.

From these three core values, eight principles were given under the Guidelines to establish a solid AI R&D environment and society that follows global AI trends, namely:

• common good and well-being;
• fairness and non-discrimination;
• autonomy and control;
• safety;
• privacy and data governance;
• transparency and traceability;
• explainability; and
• accountability and communication.

On 28 March 2023, the National Science and Technology Council (NSTC) established the Taiwan AI Center of Excellence (AICoE). According to the NSTC, the AICoE will help to streamline efforts by government agencies to formulate rules governing AI applications, and will facilitate collaborations on international projects, assist in initiating international collaborations on strengthening core AI technologies and create an AI talent pool in the Asia-Pacific region.

The Taiwan government has announced the AI Technology R&D Guidelines (see 6.1 National Standard-Setting Bodies), which reflect many of the principles discussed internationally.

Also, pursuant to the AI Basic Act, MODA shall refer to international standards or norms to promote an internationally interoperable risk classification framework for AI, and shall assist each sector-specific competent authority in formulating risk-based management regulations. In addition, sector-specific competent authorities shall formulate risk-based management regulations and assist relevant industries in formulating industry guidelines and codes of conduct (see 3.1 General Approach to AI-Specific Legislation).

As such, important standards from relevant international standard-setting bodies may be taken into account in the event of the promulgation of any AI-related law or regulations following the aforesaid principles under the AI Basic Act.

In the era of vigorous development of AI, it is not only industries that are actively promoting and introducing AI; government departments are also focusing on such technology. The Executive Yuan has announced three stages of governmental AI development, to accelerate the incorporation of AI in government departments and improve administrative efficiency:

• interface digitisation;
• internal business process digitisation; and
• field AI.

At present, AI is widely applied in government branches and the results have been fruitful, with the following examples:

• the Ministry of Finance’s Fiscal Information Agency has uncovered tax evasion with the assistance of an AI system;
• the Ministry of Justice co-operated with telecommunications operators to combine AI technology to improve prison management and turn Chiayi Detention Centre into a “smart prison”;
• the Judicial Yuan commenced the AI sentencing information system, which aims to implement fairer and more transparent sentencing by collecting decision-making data and updating data immediately so as to inform judges and citizen judges of the sentencing trends in similar cases (see 9.1 AI in the Legal Profession and Ethical Considerations for further details); and
• the government adopted AI technology in the pandemic tracking system in order to facilitate outbreak investigations and curb the infection rate of COVID-19.

It is also worth noting that the Railway Bureau conducted a trial “Smart Video Surveillance System Project” in 2019, combined with facial recognition AI, to monitor various situations on the metro and notify the station or police officers in case of special circumstances. However, the use of the facial recognition function immediately triggered privacy concerns, and there was some controversy with regard to legality. The Taiwan Railway Bureau later expressed reluctance to use this function in future.

In addition to such privacy concerns, issues such as ethics, labour replacement and transformation, and lack of legal sources will inevitably be faced by government departments when introducing AI systems.

Furthermore, it is important to note that Taiwan’s Executive Yuan introduced guidelines in August 2023 regarding how generative AI should be utilised by itself and its subordinate governmental bodies and agencies.

No judicial decisions were found in relation to government use of AI.

Please see 7.1 Government Use of AI.

Generative AI has the capability of producing sophisticated and creative content that resembles human work by learning patterns from existing data and using such knowledge to create novel works. AI cannot be considered a natural or legal person, so it cannot be considered an author (or co-author) of a work.

However, the existing data used by generative AI may be subject to certain patent rights or copyrights of a natural or legal person, potentially giving rise to tort issues. As generative AI is “self-taught” (in other words, neither the AI users nor its creators can fully understand its decision-making process), it may be difficult to determine who is liable. It remains to be seen whether judicial opinions will provide clarity on this matter in the future.

According to reliable news reports, a substantial number of court decisions have been uploaded to the public e-platform in Taiwan. The Ministry of Science and Technology of Taiwan has commissioned private sector organisations to conduct research in relation to the use of legal data analysis as a tool for compiling information on court decision patterns, predicting future decision outcomes and examining the appropriateness of court decisions. In addition, as the use of e-filing has become increasingly common in litigation practice, the Judicial Yuan has recently been contemplating establishing a “technology court” to facilitate the electronic exchange of documents and e-litigation.

It is also worth noting that the Judicial Yuan launched a sentencing system using AI technologies on 6 February 2023. This system aims to organise past sentences in order to establish a fair and transparent sentencing trend, assisting future citizen judges (the citizen judge system in Taiwan was launched on 1 January 2023) in judging cases fairly. By analysing court decisions, the system could automatically generate sentencing recommendations for the user, and provide similar cases for the judges’ and lawyers’ consideration. According to the Judicial Yuan, the system will be available to prosecutors, lawyers and the general public. The system is aimed at rapidly identifying relevant data from the vast amount of electronic data, and significantly improving the accuracy and speed of the litigation process in the near future.

In addition, using generative AI to produce legal documentation may be subject to legal issues. Even if AI may not be considered an author (or co-author) of a work, the possibility that any content (such as a draft contract or legal memo) produced by AI may still be subject to the IP rights of any third party cannot be completely ruled out.

Civil Liability

As AI itself has not yet been recognised as a legal entity, it cannot be subject to any civil liability under current Taiwan laws. For AI-related tort cases, the injured party would need to prove that the use of AI falls within the scope of tort liability stipulated under the Civil Code and/or the Consumer Protection Act, which includes (without limitation) the following.

• A manufacturer shall be liable for any damage caused by its products, unless it can prove that the products have complied with the applicable technical and professional standards of reasonably expected safety requirements before they are released onto the market (Article 7 of the Consumer Protection Act).
• If an automobile, motorcycle or other motor vehicle that does not need to be driven on tracks while in use has caused injury to another person, the driver shall be liable for the damages arising therefrom, unless they have exercised reasonable care to prevent the damages (Article 191–2 of the Civil Code).
• A person who has wrongfully infringed the rights of another person (whether intentionally or negligently) should compensate such person for any damages arising therefrom. The prevailing view among courts and scholars is that there should also be causation between the tortious conduct and the injury (Article 184 of the Civil Code).

By way of an example, if AI embedded in a self-driving system (such as an autonomous vehicle) causes an injury, the injured person may wish to prove and convince the judge that the self-driving vehicle falls within the definition of “automobile” and that the user should be considered the “driver” under Article 191–2 of the Civil Code. The injured person would also bear the burden of proving that the “user” was negligent when using the self-driving vehicle if such person wishes to establish a claim under Article 184 of the Civil Code. Furthermore, the manufacturer of the self-driving vehicle may be held liable under Article 7 of the Consumer Protection Act if the court finds that the vehicle failed to comply with the contemporary technical and professional standards of reasonably expected safety requirements before it was released onto the market.

Criminal Issues

In Taiwan, no criminal-related laws have been specifically promulgated or amended to address AI developments. As criminal liability under Taiwan law typically requires a person’s mental state to be capable of “intention” or “negligence”, AI itself would not possess such “mental state” and therefore would not be capable of committing a criminal offence. In addition, under the current Taiwan legal regime, only natural persons are deemed capable of committing crimes – apart from in certain exceptional circumstances where legal persons may be subject to criminal liability. Therefore, in order to determine whether an AI-related crime has been committed, the prosecutor would need to prove (inter alia) the intention/negligence and causation of the person “using” or “behind” the AI, in a similar manner as outlined under Civil Liability, above.

Using the above-mentioned self-driving vehicle case as an example, the prosecutor would need to prove that the vehicle “user” causing the injury acted negligently. If the user could prove that the accused activity was only the “behaviour” or “act” of AI, the court may consider that there was neither negligence on the user’s part nor causation between any act of the user and the result.

No relevant regulations have yet been proposed.

While there is currently no judicial precedent regarding agentic AI in Taiwan, the following issues are worth discussing.

AI as an “Authorised Agent”

Many are debating whether AI can be “authorised” to act as an “agent” from a legal viewpoint. Under Taiwan’s current legal framework, such an interpretation may not be feasible as AI is not a “person”. From a practical standpoint, it would require proof of whether the “actions” of an agentic AI can legally bind the user – especially in scenarios where no existing legal relationship or contractual agreement governs the rights, applications and responsibilities of the parties.

Consumer Protection and Liability

When a business offers agentic AI as a product, consumer protection considerations may be implicated. This also gives rise to liability questions: if the “actions” of an agentic AI result in harm, responsibility must be allocated – whether to the user or to the service provider. In this context, existing legal discussions on liability for self-driving vehicles may provide a useful point of reference. Please see 10.1 General Theories of Liability.

See 10.1 General Theories of Liability.

No Taiwan court decisions have addressed the issues of discrimination and bias that may be caused by the use of AI algorithms and big data analytics, nor have any specific laws or regulations been promulgated or amended to address such issues. However, as “equality” and “fairness” are important factors when it comes to social life and economic activity, more and more discussions are likely to emerge in legal fields such as labour/employment law (please see 14.1 Hiring and Termination Practices), privacy law, antitrust and other areas.

In Taiwan, facial recognition and biometric information are classed as personal information and are protected by the Taiwan Personal Data Protection Act (PDPA), under which the collection, processing and use of any personal data are generally subject to notice and consent requirements. Pursuant to the PDPA, the term “personal data” includes characteristics, fingerprints, genetic information and other biometric data and information that may directly or indirectly identify an individual.

Under the PDPA, unless otherwise specified under law, a company is generally required to give notice to and obtain consent from an individual before collecting, processing or using any of said individual’s personal information (subject to certain exemptions). As a result, if a company wishes to collect, process and/or use any personal data for purposes concerning AI and/or big data, it will be subject to the above obligations under the PDPA.

Pursuant to the AI Basic Act, the government shall avoid AI applications that:

• infringe upon the life, body, liberty or property of individuals;
• disrupt social order, national security or the ecological environment; or
• result in violations of relevant laws and regulations, such as those relating to bias, discrimination, false advertising, misinformation or fabrication.

Also, the government shall adhere to the principle of the best interests of children and adolescents.

Where an AI product or system is determined to be a high-risk application by the sector-specific competent authority in consultation with MODA, clear warnings or precautionary statements shall be explicitly indicated (please see 3.1 General Approach to AI-Specific Legislation).

Based on the above, it is expected that relevant laws and regulations will be promulgated following the enactment of the AI Basic Act. However, before such laws or regulations are formulated, legal issues arising from deepfakes and synthetic media – such as the civil liability or criminal liability – can only be addressed under the provisions of existing laws.

However, please note that, according to Taiwan’s Fraud Crime Hazard Prevention Act (FCHPA), online advertising platform operators (OAPOs) that provide advertising services in the territory of Taiwan through the internet and reach a certain scale should also comply with the provisions of the FCHPA. Relevant obligations and provisions include the following.

• Advertisements published or broadcast on online advertisement platforms may not include content involving fraud.
• OAPOs shall establish the following management measures:
1. for online advertising services, the identity of persons commissioning the publishing and broadcasting and the capital contributor shall be verified through digital signature, rapid authentication system, or other technologies or methods with equivalent safety; and
2. carrying out analysis and assessment for the risk of online advertising services being used for fraud crimes to establish a legal, necessary and effective fraud prevention plan for detection, identification and response, and publish a fraud prevention transparency report annually.
• When publishing or broadcasting advertisements on the platforms, OAPOs shall disclose the following information in the advertisements:
1. a label indicating that it is an advertisement;
2. information related to the personnel commissioning the publishing and broadcasting, and investors;
3. the licence number of the advertisement if it is legally required to have one; and
4. whether the advertisement uses deepfake technologies or AI-generated individual images.
• OAPOs that are aware that the advertisements they publish or broadcast are fraudulent or significantly involve fraud shall adhere to the following provisions.
1. Actively remove, restrict browsing and stop broadcasting such advertisements, or adopt other necessary actions within the period notified by the judiciary police department, MODA or the sector-specific competent authorities, and provide the information of the personnel commissioning the publishing and broadcasting, the investors, the network communication software accounts and telecom number in the advertisements that are suspected of involving fraud, and other relevant information, to the judiciary police department.
2. For users who publish and broadcast fraudulent advertisements or those significantly involving fraud, or users notified by the judiciary police department as accounts significantly involving fraud, the platform operator shall suspend the service provision for a reasonable period. OAPOs that violate these requirement shall bear joint responsibility for damage compensation together with the personnel commissioning the publishing and broadcasting of the advertisements and the investors for any harm caused to individuals who were misled by the content of the advertisements.
• When sector-specific competent authorities or judiciary police department notify OAPOs that the content published or broadcast on their platforms is suspected of being related to fraud, the OAPOs shall take the initiative to restrict access and browsing or remove the relevant content.

According to the AI Basic Act (see 3.1 General Approach to AI-Specific Legislation) and the AI Technology R&D Guidelines announced by the Ministry of Science and Technology under the Executive Yuan of Taiwan in 2019 (see 6.1 National Standard-Setting Bodies), transparency is an important topic.

Pursuant to the AI Basic Act, one of the principles that the government should follow in promoting the research, development, and application of AI is transparency and explainability. This principle provides that the outputs of AI should be accompanied by appropriate information disclosure or labelling, in order to facilitate the assessment of potential risks and to understand the impact on relevant rights and interests, thereby enhancing the trustworthiness of AI.

Also, the Guidelines explicitly state that, in order to ensure the fairness of the decision-making process, the development and application of AI systems, related software and algorithms (including the provision and disclosure of information on modules, mechanisms, components, parameters and calculations) must ensure that the public is informed of the elements of decisions generated by AI systems. In addition, AI development and application must comply with traceability requirements, adequate records of the decision-making process must be kept (including data collection, data labelling and algorithms used), and a record-keeping system must be established to facilitate future remedy and clarification for affected parties in AI technology decisions.

The use of AI technology comes with various risks that need to be reflected in transactional contracts between customers and AI suppliers, including but not limited to:

• privacy and data security (see 12.2 Biometric Technologies and Emotion Recognition);
• improper use (see 10.1 General Theories of Liability);
• algorithmic bias (see 12.1 Algorithmic Bias and Fairness); and
• technical failures.

To minimise potential liabilities and ensure the secure, compliant and efficient use of AI technology, businesses should:

• establish clear requirements and expectations;
• prioritise adherence to compliance and ethical standards; and
• delineate responsibility within contracts.

Prior to implementing AI technology, businesses are also recommended to establish robust business and legal frameworks, and document the testing history of algorithms to record the due diligence as well as the rationale for not using alternative solutions. Given the boundless potential and unpredictable nature of AI development and the potential legal ramifications, businesses must exercise heightened caution to mitigate the risk of substantial fines or liability for damages.

Please see 21.1 AI Governance Frameworks and Implementation.

It is understood that AI recruitment could greatly shorten the hiring process and make it more effective. AI could quickly select suitable candidates from applicants based on a series of indicators (including conditions such as work experience, education and technical skills) by using algorithms to analyse the videos, test scores, interview videos and other data submitted by the applicants. AI recruitment may also assist employers in screening applicants by analysing, for example, their facial muscle movement, vocal characteristics and other indicators.

However, certain legal issues need to be considered in terms of AI recruitment (see 14.2 Employee Evaluation and Monitoring).

In order to evaluate and monitor employee performance, employers may use employee monitoring technology (such as electronic trackers, surveillance cameras, metabolism monitors, wearable biological measuring devices and implantable technology) in the workplace to:

• predict and flag the employee’s performance;
• monitor biometric and health data of the employee to predict future healthcare costs; and
• conduct remote monitoring and time-tracking to measure the employee’s loyalty and discipline.

However, using AI to evaluate and monitor employees may lead to several issues, including invading employees’ privacy, discouraging unionisation and causing employment discrimination, thereby undermining worker protections. For more on privacy laws, please see 12.2 Biometric Technologies and Emotion Recognition; regarding labour laws, please see 14.1 Hiring and Termination Practices.

The demand for robots has been growing rapidly in recent years, and unmanned intelligent logistics has become a future trend. On 1 March 2023, the Department of Technology of the MOEA – together with a leading chain store company in Taiwan – launched an autonomous mobile robot (AMR) delivery project (“Cubot ONE”) in Kaohsiung Software Park to operate an AMR delivery service. Customers may order food directly from the mobile app, and the AMR will deliver the orders directly to customers.

According to reliable news reports, the AMR is able to accurately receive orders, perform delivery tasks, avoid pedestrians or obstacles without damaging the order, cross terrain obstacles, and call and take elevators to complete cross-floor delivery. As a result of this trend, it is likely that the demand for delivery worker jobs is likely to decline drastically in the next few years.

Please see 3.2 Jurisdictional Law for more on the law regarding autonomous vehicles/self-driving vehicles, and 14.1 Hiring and Termination Practices for further details on labour laws.

In Taiwan, AI applications (eg, chatbot and robo-adviser services) in financial services involve correspondence with clients. The Operating Rules for Securities Investment Consulting Enterprises Using Automated Tools to Provide Consulting Service (as promulgated in 2017 and last amended in 2022 by the Securities Investment Trust and Consulting Association of Taiwan, the self-disciplinary organisation of the asset management industry) provide that securities investment consulting enterprises may provide robo-adviser services ─ ie, online securities investment consulting services using automation tools and algorithms. For applicable regulations for financial services companies, please refer to 3.2 Jurisdictional Law.

In Taiwan, the Ministry of Science and Technology is driving the AI for Health Plan, which has assisted major medical research institutions in Taiwan in developing AI algorithms to be used for diagnosing cancer lesions at an early stage, accelerating image recognition, etc. Furthermore, in accordance with the 2021 Medical Device Act, the TFDA has set up a project office that aims to:

• assist with matters regarding medical devices using AI technology; and
• review the regulations to explore the possibility of amending existing rules in order to establish a better regulatory environment for AI in healthcare.

Please see 10.1 General Theories of Liability.

Please see 10.1 General Theories of Liability.

Please see 10.1 General Theories of Liability.

As AI is a research result shown by machines that are made by humans, it cannot be considered a natural or legal person, so it cannot be considered an author (or co-author) of a work. Consequently, the assets in the AI process might not be eligible for IP protection.

In Taiwan, copyright is protected by the law without registration or filing requirements. However, certain features/components should be in place in order to qualify as a copyright, such as “originality” and “expression”. Despite the fact that a “computer program copyright” is a type of copyright stipulated in the Copyright Act, only a natural person or a legal person can be classed as the author of a work to enjoy copyright upon completion of the work, according to the TIPO and recent Intellectual Property and Commercial Court (IPCC) court judgment (see 4.1 Precedent-Setting Judicial Decisions). Although AI cannot be an author (or co-author) of a work, it is still worth observing whether judicial opinions will change in the future.

Also, the assets generated through generative AI may be produced through dialogue and instructions, and due to the substantial amount of machine learning in generative AI systems, there are numerous impurities and variables between the creation and the user’s own personality. As long as the generative AI system itself does not align with the characteristics of the human spirit, the works of generative AI systems at this stage may not necessarily be eligible for copyright protection.

Although the terms and conditions of the AI tool provider may stipulate that “all content generated by the generative AI system is owned by the generative AI system”, due to the aforementioned legal issue regarding AI creation, it is worth observing whether any judicial opinions will establish that the generative AI system can claim copyright ownership over the user, and whether the aforementioned agreement is reasonable and in line with existing judicial opinions in the future.

Trade Secrets

AI cannot be considered an author (or co-author) of a work as it is not classed as a natural person or a legal person. However, seeing as there have been no law amendments in relation to AI-generated trade secrets proposed by the competent authority, the issue of whether AI is able to create a method, technique, process, formula, program, design or other information that may be used in the course of production, sales or operations under the Trade Secrets Act in Taiwan is still under observation. Such issues will be more important when AI has evolved to have the ability to think independently and create a “trade secret”. These issues might not be solved under the current trade secret regime in Taiwan and, in light of ongoing AI developments, are challenging for the government, legislators, representatives of the court system, and other legal practitioners.

In Taiwan, the question of whether or not AI can be considered an inventor on a Taiwan patent has been widely discussed in the legal profession in recent years. According to the IPCC of the TIPO, creation is a collective term for the fruits of human mental activity, which may be protected with different types of rights depending on the content and nature of the creation.

In respect of the right of attribution, the Enforcement Rules of the Patent Act also provide that the application form must specify the name and nationality of an inventor, utility model creator or designer when a patent application is filed. According to the provisions of the relevant laws and regulations in Taiwan, the inventor must be not only a person who has made substantial contributions to the technical features of the patent application, but also a natural person. Given that an AI system is still regarded as a “thing” under the laws of Taiwan, it is therefore not entitled to enjoy legal capacity and natural personality.

The Supreme Administrative Court further pointed out that an inventor’s right of attribution is a personality right, so the inventor must be a natural person, which is in line with the legislative intent of the Patent Act and associated regulations. Where a thing that is not a natural person is specified as an inventor in the application documents, the application is not deemed to comply with the statutory forms and processes – although it is still possible to make a correction in this respect. It is nonetheless worth observing whether judicial opinions will change in the future.

According to the Copyright Act, AI-generated works may still infringe upon the copyrights of others even if the user of such works is not involved in profit-making activities. The terms of use of the AI tool provider and the provisions of the media platform regarding public broadcasting and reprints shall also be taken into consideration.

Based on the current prevailing view, the scope of IP protection under relevant Taiwan laws and regulations does not include works of art and works of authorship generated by AI.

In Taiwan, there has not yet been extensive discussion or a clear consensus on the legal complexities surrounding open-source AI. Key issues for further consideration may include the following.

• The legal rights and obligations are generally governed by service agreements or terms of use. Providers may retain ownership of the model and its parameters, while granting users a limited licence to use it.
• Providers may assign ownership of AI-generated outputs to users or recognise them as the users’ property. At the same time, terms may allow providers to use users’ inputs and outputs for model training or improvement.
• When AI-generated content infringes IP rights, liability allocation remains unclear. Providers may argue that infringement stems from user-supplied training data or parameter settings, potentially shifting responsibility to the user.
• Users’ rights to use AI-generated content, including for commercial purposes, may be restricted by the provider’s authorisation.
• Since AI is not considered a “person”, users may face challenges claiming copyright protection under existing law, even if their contract allows commercial use.

Pursuant to the AI Basic Act, the sector-specific competent authority – in consultation with the competent authority for personal data protection – shall avoid any unnecessary collection, processing or use of personal data during the research, development and application of AI, and shall promote the incorporation of personal data protection into default and design-related measures or mechanisms to safeguard the rights and interests of data subjects (see 3.1 General Approach to AI-Specific Legislation). However, before any relevant amendments are made to the PDPA, the original provisions shall still apply.

Pursuant to the PDPA, the collection, processing and use of any personal data are generally subject to notice and consent requirements. Under the PDPA, the term “personal data” includes characteristics, fingerprints, genetic information and other biometric data, and information that may directly or indirectly identify an individual.

Under the PDPA, unless otherwise specified under law, a company is generally required to give notice to and obtain consent from an individual before collecting, processing or using any of said individual’s personal information (subject to certain exemptions). As a result, if a company wishes to collect, process and/or use any personal data for purposes concerning AI and/or big data, it will be subject to the above obligations under the PDPA.

See 17.1 AI Training and Data Protection.

See 17.1 AI Training and Data Protection.

Pursuant to Taiwan’s Fair Trade Act (FTA), a “concerted action” (ie, so-called cartels) means that competing enterprises at the same production and/or marketing stage jointly determine – by means of “contract, agreement or any other form of mutual understanding” – the price, technology, products, facilities, trading counterparts or trading territory in respect of goods or services, or any other behaviour that restricts each other’s business activities, thereby resulting in an impact on the market function with regard to production, trade in goods or supply and demand of services. According to the FTA, the term “any other form of mutual understanding” means “a meeting of minds” – whether legally binding or not – that would in effect lead to joint actions.

If the competing enterprises’ actions are taken by AI, there may be an issue regarding whether these price-setting actions are indeed caused by “any other form of mutual understanding” among the enterprises if no explicit contract or agreement exists among the companies. Therefore, whether or not the firms have a “meeting of minds” could be an issue when debating in Taiwan court.

According to the AI Basic Act, information security and safety is one of the principles with respect to the government’s promotion of AI R&D and application (see 3.1 General Approach to AI-Specific Legislation).

In January 2025, based on considerations of national information and communication security, MODA issued a special warning to government agencies and critical infrastructure to restrict the use of DeepSeek AI products. This measure aims to prevent user-related data or information being transmitted by products with information security concerns, which could pose a threat to national information and communication security. MODA stated that DeepSeek AI services are Chinese products, and their operation involves cross-border transmission and information leakage, raising information security concerns. These products are considered a threat to national information and communication security.

To prevent the leakage of internal information from government agencies under conditions where effective supervision is not possible, and to mitigate associated risks, since 2019 the Executive Yuan of Taiwan has implemented the “Principles for Restricting the Use of Products Harmful to National Information and Communication Security”, which clearly require central and local government agencies, public schools, state-owned enterprises, administrative legal persons and venues operated or outsourced for public activities or use to restrict the use of products harmful to national information and communication security.

AI can theoretically identify potential ESG risks by analysing large amounts of data, allowing for more efficient management of these issues and helping companies reduce problems and risks. In practical applications, using AI technology for prediction and modelling can assist companies in developing more effective risk management strategies. In Taiwan, there are no specific regulations regarding the impact of AI on ESG, nor are there restrictions on its use. However, when companies use AI tools, management must still pay attention to governance and fiduciary duty-related issues (see 21.1 AI Governance Frameworks and Implementation) to ensure a certain level of quality and reliability, whether the AI is developed internally or procured externally.

It is crucial for companies to be aware of data protection laws, IP laws and labour laws when using AI technologies. For further details, please refer to 7.1 Government Use of AI, 12.2 Biometric Technologies and Emotion Recognition, 14. Employment and 16. Intellectual Property.

Regarding governance, the Taiwan Company Act has stipulated the director’s fiduciary duty and the obligation to act in good faith. A director of a company must be loyal and exercise the due care of a good administrator when conducting the business operations of the company or else the director will be liable for the loss or damage thereby sustained by the company. The aforementioned standards of “loyalty” and “due care of a good administrator” when conducting the business operations of a company, in general, should be based on the actual circumstances, according to objective and socially recognised criteria. For instance, the board may consider referencing and relying on experts’ views and opinions before making any decisions at the board level, thereby reducing the risk of potential breach of fiduciary duty claims.

The same principle applies in AI-related cases. There is currently no explicit court precedent and ruling in this respect. However, if the directors are not experts in AI-related fields, they are advised to engage an AI expert to provide further advice when making AI-related decisions. Directors may fulfil their fiduciary duty by engaging outside AI experts and, at the same time, obtain a solid basis to support the legitimacy of their decisions.

In addition, companies that wish to use AI technologies for business prediction or risk assessment should carefully identify the relationship between the input and output of the AI data processing, including how the AI processes the data and its underlying logical relationships. Therefore, the board of directors will be able – with the assistance of external experts – to identify the underlying AI algorithms, and preventative measures can be taken to avoid the possible failure and inaccuracy of the AI system.