夜雨聆风Platforms urged to enforce pre-reviews and traceability to counter AI face theft
Illustration: Liu Xiangya/GT
"AI does not understand that sometimes silence speaks louder than shouting, and imperfection is perfect…That one percent injection of 'lived-in soul' is what makes AI works stand out," wrote Feng Shengyong, director of the TV Drama Department of the National Radio and Television Administration (NRTA), in an article published by the People's Daily on Tuesday.
While acknowledging the efficiency revolution brought by AI, Feng also warns of the risk of "disorder" as AI-generated contentbecomes increasingly indistinguishable from reality, making the protection of personal rights and interests an urgent issue.
Experts noted that the solution lies in clearly strengthening platforms' pre-review and material traceability obligations, andshifting the burden of proof to content creators.
The urgency of this issue has been highlighted by a spate of high-profile AI face theft cases. On Sunday, Chinese star Yi Yangqianxi, also known as Jackson Yee, released a statement condemning the unauthorized use of his likeness in AI short dramas on multiple platforms and demanding their immediate removal.
A day later, Hongguo, a free short drama streaming platform, issued an announcement about governing the illegal use of materials in AI short dramas. The announcement stated that in the first quarter of 2026, the platform had removed 1,718 AI-generated comic-style short dramas that violated its governance norms. Moreover, the platform launched a special centralized rectification campaign that completed comprehensive inspections of 15,000 works, disposed of 670 illegal works in accordance with regulations, and exposed four types of typical cases, the China News reported on Tuesday.
Earlier, a Hanfu (a type of traditional clothing of the Han ethnic group) enthusiast became a victim of face theft by an AI short drama in which his image was deliberately defamed and spread on platforms such as Hongguo, sparking a trending topic on China's X-like platform Sina Weibo.
From top celebrities like Yee, Xiao Zhan, Yang Zi and Dilraba Dilmurat, whose likenesses have been copied to drive traffic for short dramas, to even ordinary people reduced to "raw materials" in the AI content industry chain, AI face theft has spread from celebrities to the general public.
Alongside face theft, "voice theft" is also rampant. More than 20 dubbers recently issued a joint statement on social media, demanding an end to the unauthorized collection of their voices for AI training and commercialization.
In early April, the Actors Committee of the China Federation of Radio and Television Associations spoke out against such infringements, marking the official start of a "face protection conflict" concerning everyone's digital personality.
Zhu Xinmei, director of the Institute of International Communication at the Development Research Center of the NRTA, noted that theworseningstate of AI face theft is rooted in the sharp lowering of technical thresholds and the lag in regulatory governance.
"AI face-swapping technology has completely moved beyond professional fields to become popular and automated," she told the Global Times on Tuesday. "The convenience of generating face-swapped content with one click allows anyone to complete face theft in a short time, while the speed of AI content generation far exceeds the capacity of manual review."
She added that platforms' excuse of "being unable to keep up with the speed of content production" essentially stems from a lack of both technical prevention and responsibility.
The core driver behind the persistent infringements, Zhu argues, is the serious imbalance between infringement costs and benefits.
Infringers avoid the high cost of hiring celebrities by stealing their faces, while repeating substantial online benefits. This creates a vicious cycle that is difficult to break.She listed an example that although the Beijing Internet Court established the "identifiability" principle for infringement in Dilraba's case,the long cycle and high cost of judicial rights protection mean most infringers have already profited before being held accountable.
Legal ambiguity further exacerbates the problem. Wu Xiaolin, a Beijing-based lawyer, pointed out thatthe most pressing difficulty in reconstructing rules concerning image rights in the AI era is the "permanent digital identity occupation" caused by face theft.
"Without the technical means to block it, using legal measures to cut off the source of digital identity infringement will be the key breakthrough," he told the Global Times on Tuesday.
Wu suggests clarifying the "identifiability" standard in the law by setting a 60 percent similarity threshold.Under this standard, any AI-generated face that can be linked to a specific individual should constitute infringement.
He also proposed increasing the evidence weight of victim-provided materials and promoting electronic evidence preservation platforms to ease the burden of proof.
To addressthese issues, Zhu emphasized the need for coordinated efforts from the legal framework, platforms and technology. She called for stronger legal regulations to clarifythe protection of personal data assets and standards for AI-generated faces. She also urged platforms to establish technical preventionsystems to intercept infringement at the source, while promoting digital watermarking and traceability mechanisms. Echoing director Feng's proposal to "focus on both standardized management and application improvement," both experts agreed that shifting the burden of proof to creators is crucial. That is, creators - rather than victims- should be required to prove data compliance.
平台被敦促加强预审与可追溯性以应对AI换脸盗用
“AI并不理解，有时沉默比呐喊更有力量，不完美才是完美……那百分之一的‘生活过的灵魂’的注入，才是让AI作品脱颖而出的关键，”国家广播电视总局电视剧司司长冯胜勇在周二发表于《人民日报》的一篇文章中写道。
在承认AI带来的效率革命的同时，冯也警告说，随着AI生成的内容变得越来越难以与现实区分，“失序”风险随之而来，这使得保护个人权益成为一个紧迫的问题。
专家指出，解决方案在于明确加强平台的预审和素材可追溯性义务，并将举证责任转移给内容创作者。
这一问题的紧迫性已被一系列备受瞩目的AI换脸盗用案件所凸显。周日，中国明星易烊千玺发表声明，谴责多个平台上未经授权将其肖像用于AI短剧的行为，并要求立即删除这些内容。
一天后，免费短剧流媒体平台红果发布了关于治理AI短剧中非法使用素材的公告。该公告称，2026年第一季度，该平台已下架了1718部违反其治理规范的AI生成漫画风格短剧。此外，据中国新闻社周二报道，该平台发起了专项整治行动，完成了对15000部作品的全面检查，依规处置了670部违规作品，并曝光了四类典型案例。
更早之前，一位汉服爱好者在AI短剧中成为换脸盗用的受害者，他的形象被故意丑化并在红果等平台上传播，引发了国内类似X平台的新浪微博上的一个热门话题。
从易烊千玺、肖战、杨紫、迪丽热巴等顶级明星（他们的肖像被复制来为短剧引流），到甚至普通人也沦为AI内容产业链中的“原材料”，AI换脸盗用已从名人蔓延至普通大众。
除了换脸，“声音盗用”也很猖獗。 20多位配音演员近日在社交媒体上发布了一份联合声明，要求停止未经授权收集他们的声音用于AI训练和商业化。
四月初，中国广播电视社会组织联合会演员委员会发声反对此类侵权行为，标志着一场关乎每个人数字人格的“面容保卫战”正式打响。
国家广播电视总局发展研究中心国际传播研究所所长朱新梅指出，AI换脸盗用日益恶化的现状根源于技术门槛的急剧降低和监管治理的滞后。
“AI换脸技术已经完全走出专业领域，变得普及化和自动化，”她周二告诉《环球时报》。“一键生成换脸内容的便捷性让任何人都能在短时间内完成换脸盗用，而AI内容生成的速度远远超过人工审核的能力。”
她补充说，平台“跟不上内容生产速度”的借口，本质上源于技术防范和责任心两方面的缺失。
朱认为，侵权屡禁不止的核心驱动因素，是侵权成本与收益之间的严重失衡。
侵权者通过盗用名人面容规避了聘请名人的高额成本，同时在网上获取大量收益。这就形成了一个难以打破的恶性循环。 她举了一个例子：虽然北京互联网法院在迪丽热巴案中确立了侵权的“可识别性”原则，但司法维权的周期长、成本高，意味着大多数侵权者在被追责之前就已经获利。
法律上的模糊性进一步加剧了问题。 北京律师吴晓林指出，“在AI时代重构肖像权规则时，最紧迫的难点在于换脸盗用造成的‘数字身份永久占用’。”
“在没有技术手段进行拦截的情况下，通过法律手段切断数字身份侵权的源头将是关键突破口，”他周二告诉《环球时报》。
吴建议在法律中明确“可识别性”标准，设定60%的相似度阈值。在这一标准下，任何能够关联到特定个人的AI生成面孔都应构成侵权。
他还提议提高受害者提供材料的证据效力，并推广电子证据存证平台，以减轻举证负担。
为了解决这些问题，朱强调需要法律框架、平台和技术协同发力。她呼吁加强法律法规，明确个人数据资产的保护以及AI生成面孔的标准。她还敦促平台建立技术防范系统，从源头上拦截侵权行为，同时推动数字水印和可追溯机制。呼应冯胜勇司长“规范管理和应用提升两手抓”的提议，两位专家一致认为，将举证责任转移给创作者至关重要。也就是说，应由创作者——而非受害者——来证明数据合规性。
【语法填空改编短文】
AI face theft, also known as face-swapping abuse, has become a serious issue. With AI technology making generated content increasingly indistinguishable from reality, personal rights and interests are facing unprecedented threats. From celebrities to ordinary people, many have fallen victim to unauthorized use of their likenesses in AI-generated short dramas. Even voice theft is rampant, with over 20 dubbers recently demanding an end to the commercialization of their voices without permission.
Some streaming platforms have taken action. For instance, Hongguo, a free short drama platform, removed over 1,700 illegal AI works in the first quarter of 2026 and conducted a comprehensive inspection of 15,000 works. However, the worsening situation is mainly due to the low technical threshold and lagging regulatory governance. AI face-swapping has become automated and popular, allowing anyone to complete face theft with one click. Meanwhile, the speed of AI content generation far exceeds manual review capacity.
To solve the problem, experts propose several measures. First, platforms should strengthen pre-review and traceability obligations, shifting the burden of proof from victims to content creators. Second, legal standards need to be clarified. A 60% similarity threshold for identifiability has been suggested, meaning any AI-generated face that can be linked to a specific individual should constitute infringement. Third, technical prevention systems, such as digital watermarking and traceability mechanisms, should be established to intercept infringement at the source. Only through coordinated efforts from legal frameworks, platforms, and technology can this vicious cycle be broken.
AI换脸，也称为换脸滥用，已经成为一个严重的问题。随着AI技术使得生成内容越来越难以与现实区分，个人权益正面临前所未有的威胁。从名人到普通人，许多人已成为未经授权在AI生成的短剧中使用其肖像的受害者。甚至声音盗用也很猖獗，最近有20多位配音演员要求停止未经许可将其声音商业化。
一些流媒体平台已经采取行动。例如，免费短剧平台红果在2026年第一季度下架了1700多部违规AI作品，并对15000部作品进行了全面检查。然而，日益恶化的局面主要是由于技术门槛低和监管治理滞后。AI换脸已经变得自动化和普及化，任何人都可以一键完成换脸。同时，AI内容生成的速度远远超过人工审核能力。
为了解决这个问题，专家提出了几项措施。首先，平台应加强预审和可追溯性义务，将举证责任从受害者转移到内容创作者。其次，法律标准需要明确。已经提出了60%的可识别性相似度阈值，这意味着任何能关联到特定个体的AI生成面孔都应构成侵权。第三，应建立技术预防系统，如数字水印和可追溯机制，从源头上拦截侵权。只有通过法律框架、平台和技术的协同努力，才能打破这个恶性循环。
【词汇解释】
1. platformn. 平台；讲台；纲领
Streaming platforms are required to strengthen their pre-review of AI-generated content.
流媒体平台被要求加强对AI生成内容的预审。
2. traceabilityn. 可追溯性；可追踪性
Traceability mechanisms help identify the source of illegal face-swapped videos.
可追溯机制有助于识别非法换脸视频的来源。
3. indistinguishableadj. 无法区分的，难以辨别的
AI-generated faces are becoming indistinguishable from real human faces.
AI生成的面孔正变得与真实人脸难以区分。
4. urgencyn. 紧迫性；急迫
The urgency of protecting personal image rights has been highlighted by recent face theft cases.
近期的换脸侵权案件凸显了保护个人肖像权的紧迫性。
5. infringementn. 侵权；违反
Using a celebrity's face in an AI drama without permission constitutes an infringement.
未经许可在AI短剧中使用明星面孔构成侵权。
6. unauthorizedadj. 未经授权的；未经许可的
The unauthorized collection of voices for AI training violates personal rights.
未经授权收集声音用于AI训练侵犯了个人权利。
7. commercializev. 商业化；使商品化
Companies should not commercialize personal data without the owner's consent.
未经所有者同意，公司不应将个人数据商业化。
8. regulatoryadj. 监管的；管理的
Regulatory governance needs to catch up with the rapid development of AI technology.
监管治理需要跟上AI技术的快速发展。
9. identifiabilityn. 可识别性；可辨识性
A 60% similarity threshold for identifiability could help define AI face infringement.
60%的可识别性相似度阈值有助于界定AI换脸侵权。
10. burden of proofn. 举证责任
Shifting the burden of proof to content creators makes it easier for victims to protect their rights.
将举证责任转移给内容创作者，能让受害者更容易维护自己的权利。
【语法填空改编】
AI face theft, also known as face-swapping abuse, has become a serious issue. With AI technology making generated content increasingly indistinguishable 1.______reality, personal rights and interests are facing unprecedented threats. From 2._______(celebrity)to ordinary people, many have fallen victim to unauthorized use of their likenesses in AI-generated short dramas. Even voice theft is rampant, with over 20 dubbers recently demanding 3.______end to the commercialization of their voices without permission.
Some streaming platforms have taken action. For instance, Hongguo, a free short drama platform, removed over 1,700 illegal AI works in the first quarter of 2026 and conducted a comprehensive inspection of 15,000 works. However, the worsening situation is mainly due to the low technical threshold and 4._______(lag)regulatory governance. AI face-swapping 5.______(become)automated and popular, allowing anyone to complete face theft with one click. Meanwhile, the speed of AI content generation far 6.______(exceed)manual review capacity.
To solve the problem, experts propose several measures. First, platforms should strengthen pre-review 7.______traceability obligations, shifting the burden of proof from victims to content creators. Second, legal standards need to be clarified. A 60%8._______(similar)threshold for identifiability has been suggested, meaning any AI-generated face that canbe linkedto a specific individual should constitute infringement. Third, technical prevention systems, such as digital watermarking and traceability mechanisms, should 9.______(establish)to intercept infringement at the source. Only through 10.______(coordinate)efforts from legal frameworks, platforms, and technology can this vicious cycle be broken.
【参考答案】
1.from 2.celebrities 3.an 4.lagging 5.has become 6.exceeds 7.and 8.similarity 9.be established 10.coordinated
【答案详解】
第1题
答案： from
解析：本题考查固定搭配。indistinguishable 意为“无法区分的”，常与介词 from 连用，构成 be indistinguishable from...，表示“与……难以区分”。句中指AI生成的内容与现实难以区分。
第2题
答案： celebrities
解析：本题考查名词复数。celebrity 是可数名词，此处泛指“名人”这一类人，且与后面的 ordinary people（普通人）并列，应用复数形式 celebrities。
第3题
答案： an
解析：本题考查冠词。end 在这里是可数名词单数，表示“结束”，前面需要不定冠词。由于 end 以元音音素 /e/ 开头，故用 an。固定短语 put an end to（终止……）也可辅助记忆。
第4题
答案： lagging
解析：本题考查非谓语动词作定语。lag 意为“滞后”，此处修饰 regulatory governance（监管治理），表示“滞后的监管治理”。逻辑上是主动关系，故用现在分词 lagging。相当于 regulatory governance that lags behind。
第5题
答案： has become
解析：本题考查动词时态。根据上下文，AI换脸技术“已经变得”自动化和普及化，强调过去发生的动作对现在造成的影响（现在人人都能一键换脸），应用现在完成时。主语 AI face-swapping 为单数，故填 has become。
第6题
答案： exceeds
解析：本题考查主谓一致和一般现在时。句子陈述客观事实（AI生成速度远超人工审核能力），用一般现在时。主语 the speed 为第三人称单数，谓语动词加 -s，填 exceeds。far 是副词，修饰 exceeds，表示“远远超过”。
第7题
答案： and
解析：本题考查并列连词。pre-review（预审）和 traceability obligations（可追溯性义务）是平台需要加强的两个方面，两者为并列关系，应用 and 连接。
第8题
答案： similarity
解析：本题考查词性转换。空格前有 60%，后有 threshold（阈值），中间需要名词作定语或构成名词短语。similar 是形容词，其名词形式为 similarity，表示“相似度”。a 60% similarity threshold 意为“60%的相似度阈值”。
第9题
答案： be established
解析：本题考查情态动词的被动语态。technical prevention systems（技术预防系统）与 establish（建立）之间是被动关系，即“系统被建立”。should 后接动词原形，故填 be established。
第10题
答案： coordinated
解析：本题考查非谓语动词作定语。efforts（努力）与 coordinate（协调）之间是被动关系，表示“被协调的努力”，即多方协同的努力。过去分词 coordinated 作定语，修饰 efforts。类似用法：a coordinated attack（协同攻击）。
