◎ Chen Xiuan
AI technology is advancing rapidly, allowing the cost of forging images to drop rapidly. Therefore, the Executive Yuan proposed a draft amendment to the "Criminal Code" to injure the reputation of others, adding provisions for aggravated penalties for those who use the Internet or deepfake technology to infringe on the reputation of others. (File photo)
Generative AI has allowed the cost of forging images to drop rapidly. A photo or a short piece of sound can be made into content that the person involved has never said or done. The Executive Yuan proposed a draft amendment to the "Criminal Code" to injure the reputation of others, adding provisions for aggravated penalties for those who use the Internet or deepfake technology to infringe on the reputation of others. This amendment has its practical background. Traditional slander may be a paragraph of text or a rumor, but false content can create a highly realistic "sense of evidence" and spread much faster than in the past. However, once criminal law intervenes in the field of speech, the problem cannot just stop at "AI fraud is abominable." Which deepfake content should be punished and which is still satire, creation or public commentary must still be demarcated through clear criminal elements. Otherwise, while cracking down on technological infringement, it may also expand the scope of the state's punishment of speech.
What makes the crime of honour particularly difficult is that it deals with a conflict between two constitutional rights. On the one hand, the reputation and personality of the people are protected by law; on the other hand, freedom of speech is an important basis for the formation of public opinions in a democratic society. The Judicial Yuan has repeatedly faced this conflict in the past when dealing with defamation crimes. Now with the addition of deepfake technology, the legal arguments are even more complicated. Suppose someone makes an AI video of politicians singing and dancing. Although the scene is completely fictitious, viewers clearly know that it is a parody. This is obviously different from making a highly realistic video of bribery and deliberately making society believe that it is a real video. The legal evaluation is obviously different. The factors that the criminal law should really focus on should include whether the content is enough to cause ordinary people to mistake it for truth, whether the producer has the intention to deceive, and whether the content is enough to detract from the social evaluation of a specific person. The mere use of AI or deepfake technology should not directly become a sufficient condition for the establishment of a crime.
The photo of Trump and his aide Natalie Harp kissing passionately on a golf course is a deepfake photo created by AI. If you look closely, you can see that Harp has six fingers. (Extracted from X)
This also involves the "last resort" aspect of criminal law. Penalty is the country’s strongest legal sanction, which can leave a criminal record or even deprive one’s personal freedom. Therefore, it should usually be reserved for behaviors that infringe legal interests to a considerable extent. There are also degrees of harm caused by deepfake content: the creation of an obviously joke picture between friends, the non-consensual use of celebrity likenesses in commercials, and the deliberate production of videos of crimes, sexual acts or bribery to destroy the reputation of others. They cannot all be subject to the same criminal law evaluation just because they use generative AI. If the legislation wants to increase penalties, it should also clarify whether the basis for the increase comes from the scale of communication, the verisimilitude of the content, the results of the infringement, or the perpetrator's use of technology to increase the effect of deception. After the penalty is increased, the constituent elements must also be more precise. This is exactly what the principle of legality of crime and punishment and legal clarity require of criminal legislation.
Another more practical difficulty is proof. In future criminal cases, if you see a suspected AI video, you will not be able to directly infer that the author has intentional defamation. Prosecutors still have to prove that the defendant's behavior complies with the elements of the crime, including who produced the content, who distributed it, whether it was directed to a specific victim, and whether the perpetrator subjectively knew the content was false but still intentionally spread it. Generative AI may have been modified, downloaded, and forwarded by multiple people, and the original files may even be located on overseas platforms. The source, integrity, and identity of electronic evidence may become issues in court. What really needs to be strengthened in future practice is digital forensics, platform data preservation and electronic evidence investigation capabilities. No matter how heavy-handed the criminal law is, if it is impossible to prove who produced it, who distributed it, and whether the file was modified, the threat may still remain on paper in the end.
The more practical problem faced by generative AI for forging videos is “proof”. In future criminal cases, if you see a suspected AI video, you will directly infer that the author has intentional defamation; schematic diagram. (AFP file photo)
AI deepfakes have indeed changed the scale and methods of reputation infringement. It is necessary to discuss the adjustment of criminal law in response to technological development. However, the quality of legal revision cannot be measured solely by the degree of punishment. More importantly, "deep fake technology" should be transformed into legal standards that can be operated by the courts, distinguishing malicious deception, public comments, parody creation and general reposting, while retaining the review of the perpetrator's subjective intention and evidentiary ability. The more technology can create content that is difficult to distinguish between true and false, the more precise the law needs to be. Otherwise, the biggest problem in the future may become that we know that the criminal law punishes deep hypocrisy, but we don’t know what kind of deep hypocrisy should be jailed. Criminal law can respond to new technological risks, but in the end it should still return to the most traditional principles: before punishing a person, the state must clearly define the boundaries of the crime and must produce sufficient evidence to prove the crime.
