Although Taiwan's AI Basic Act came into force on January 14, 2026, it mainly sets forth a framework of organizational and policy initiatives that government agencies shall implement. It does not address the use of AI by courts, litigants, attorneys, or witnesses in court proceedings. Many of the procedural and evidentiary issues arising from AI's growing role in litigation remain matters of ongoing development and exploration.
Restriction on the Use of AI
However, some judges do have felt the need to draw a line regarding the use of AI in court proceedings. In a domestic violence case decided by the Taoyuan District Court in April 2026, the defendant was ordered to write a statement of remorse consisting of at least 2,000 Chinese characters, which, the judgment stresses, “must be handwritten personally and must not be drafted using an AI software.” 114 Shen-Jian-2294, Taoyuan District Court (Apr. 2026).
Use of AI by Courts and Prosecutors
In other contexts, however, courts have shown a more receptive attitude toward the use of AI. For example, in a tort action involving public bus services, the court, acting on its own initiative, used ChatGPT to calculate taxi fares between the plaintiff's residence and medical facilities and relied on those calculations in finding the plaintiff's claimed transportation expenses during rehabilitation to be reasonable. 113 Jian-Shang 65, New Taipei District Court (Oct. 2025).
In a dispute concerning the sale and purchase of a farm, the court used ChatGPT to examine fluctuations in pig feed prices over a period of time as a reference in determining whether the buyer had suffered damages from being unable to obtain the disputed farm for pig-raising purposes. 112 Shang-Geng(1) 27, Taiwan High Court Kaohsiung Branch Court (June 2024).
In a criminal defamation case, the court accepted the prosecutor's use of AI-generated evidence to rebut the defendant's claim that his online posts damaging the complainant's sexual reputation had merely been “randomly generated” by ChatGPT in response to neutral prompts such as “love,” “sensational,” and “soap opera,” and that he had published the posts only as an experiment to test readership traffic. The prosecutor's rebuttal consisted of ChatGPT's refusal to generate passages containing the sexually explicit keywords appearing in the disputed post, together with ChatGPT's usage policies. These exhibits were treated by the court as supporting evidence that the defendant's account was untrue. 113 Yi 383, Nantou District Court (Sep. 2024).
Admissibility & Evidentiary Weight of AI-Generated Content
Taiwan courts have also issued a number of decisions carefully scrutinizing both the admissibility and evidentiary weight of AI-generated content. For example, in a telecommunications fraud case involving the use of facial-recognition technology to determine whether the person appearing in an ATM surveillance footage was the defendant, the Supreme Court, in its remand judgment, indicated:
Facial recognition is a biometric identification technology based on artificial intelligence.... Current facial-recognition technology remains susceptible to environmental factors (such as lighting, angles, and obstructions), the accuracy of algorithms, and even the deepfake technology. It is also subject to the “black box effect” which means that, because of the complexity and opacity of AI algorithms, users are often unable to understand the decision-making process or how conclusions are reached, thereby limiting its verifiability.
In criminal proceedings, when addressing the admissibility of such evidence, courts should consider the nature of the evidence and the circumstances of the particular case in determining whether the hearsay rule and its exceptions apply. Courts should also consider whether other reliable evidence exists to provide corroboration. Facial-recognition matching results alone may not serve as the sole basis for identifying the defendant as the perpetrator. 114 Tai-Shang 5166, Supreme Court (March 2026). (Highlights added; same as below.)
In another case involving a former employee accused of unauthorized access to a former employer's computer system, the Taoyuan District Court dismissed an AI-generated response obtained by the complainant company regarding cybersecurity issues. 113 Yi 1166, Taoyuan District Court (Dec. 2025). While Gemini AI asserted that server log records, although editable, would leave metadata traces if altered, the court was not persuaded and stated that:
Whether content or responses generated by a multimodal large language model are accurate depends in part on the user's prompts and on the datasets employed in training the model. Where a response fails to provide a reasonable explanation or identify its sources, a court cannot simply accept the model's output at face value.
Queries to AI as Evidence
In a sexual misconduct case, the key evidence was questions that a junior-high-school student posed to ChatGPT, asking whether certain conduct by his living technology teacher done to him was normal. The conversation record was later discovered by the victim's mother, who saved screenshots of the exchange before the original record was reportedly deleted during the pre-suit gender equity investigation. The Taoyuan District Court held:
The screenshots of the ChatGPT conversations in this case constitute digital traces preserved from online interactions between witness A (the victim) and ChatGPT after the incident. The records were objectively generated and stored as electronic data through the operation of a computer system and were subsequently captured in screenshot form. They therefore constitute non-testimonial evidence....
The contents of the screenshots objectively record the questions posed by witness A (the victim) after the incident and bear a natural connection to the facts of this case. Moreover, there is no affirmative evidence in the record indicating that the screenshots were obtained through unlawful procedures or were fabricated or altered by human intervention so as to warrant exclusion. Accordingly, the screenshots possess evidentiary admissibility. 114 Ching-Su 182, Tauyuan District Court (May 2026).
Perhaps even more noteworthy is the question of whether a defendant's interactions with an AI chatbot may be admitted as evidence. The author has not yet found any Taiwanese judgments that directly address this issue. Nevertheless, at least technically, obtaining a defendant's AI chat logs may not be entirely impossible in Taiwanese criminal proceedings. Whether Taiwanese courts will eventually have to address disputes similar to those seen in the US federal cases of Warner v. Gilbarco, Inc. and United States v. Heppner remains to be seen.
Concerns about AI Changed Existing Judicial Practices
Finally, the fact that AI can itself be used as a tool of wrongdoing has prompted courts to reconsider certain existing judicial practices. In recent years, a number of courts have adopted a more restrictive approach toward requests by parties for copies of courtroom audio recordings, particularly in family-law matters. In as early as July 2022, the Taiwan High Court has observed:
Given the nature of modern digital files, which can be easily copied and rapidly disseminated, once courtroom recordings leave the control of the courts, it is technically impossible to prevent malicious actors from using various forms of deepfake technology to synthesize personal information such as voiceprints and create false audio/video content. Coupled with the speed of online dissemination, the resulting harm to the affected individual may far exceed imagination and constitute a serious infringement of personality rights....
Although the Court Organization Act does provide for monetary penalties, such sanctions are merely ex post remedies and cannot prevent further copying or dissemination of courtroom recordings by third parties. 111 Sheng 39, Taiwan High Court (July 2022).
On a final note, while the US has seen a number of cases in which lawyers were sanctioned for submitting briefs containing AI-generated hallucinated citations, no comparable cases have yet surfaced in Taiwan. This may be attributable to Taiwan's civil-law tradition, in which judicial precedents generally play a less important role in litigation than they do in common-law countries. Nevertheless, judicial sanctions would certainly be possible should a similar incident arise before a Taiwanese court.