当AI开始写歌：音乐保护的边界变了吗？

4月26日是世界知识产权日，今年的主题与"音乐"有关。在Suno.com等平台上，仅需输入故事片段并选定风格，一首完整动人的歌曲便能在数秒间流淌而出。惊叹于技术之迅猛发展之余，一个问题浮出水面：AI创作的音乐应如何界定保护？能否被纳入保护之列？

26 April marks World Intellectual Property Day, with this year’s theme focusing on "music". Platforms like Suno.com now allow users to generate complete, polished songs in seconds by inputting a story snippet and selecting a style. Amidst this rapidly advancing technology, a critical question arises: How should AI-generated music be legally protected, and is it eligible for protection at all?

独特的音乐片段—通常是简短鲜明的音频标识—可变成商标获得"无限可续期的"保护。经典如二十世纪福斯公司的电影开头曲[1]以及2016年成为中国首例注册成功的声音商标—中国国际广播电台开场曲[2]等等。AI生成的音乐虽可尝试进行商标注册，但成功门槛极高：需规避功能性声响（如闹钟提示音），更须具备鲜明品牌联想性，这对多数AI创作者而言遥不可及。

Distinctive musical segments - typically short, iconic audio identifiers - can be transformed into trade marks to obtain "perpetually renewable" protection. Classics include the Twentieth Century Fox films’ opening theme [1], as well as China’s first successfully registered sound trade mark in 2016, the opening theme of China Radio International [2], among others. While AI-generated music could theoretically seek trade mark registration, the bar is prohibitively high: the music must avoid functional sounds (e.g., alarm beeps) and demonstrate strong brand association - a hurdle most AI creators cannot clear.

音乐作品的主流护盾仍是著作权。尽管保护期非永恒，但五十年起的期限已足够长久。各国法律普遍以"独创性"与"人类智力成果"作为著作权根基。

Copyright remains the primary shield for musical works. Though protection is not perpetual, its minimum 50-year term is already substantial. Globally, copyright hinges on two pillars: "originality" and "human intellectual creation."

人类创作作品的争议多围绕"独创性"展开，常与抄袭指控交织。例如，英国歌手Ed Sheeran的《Thinking Out Loud》因被指与美国歌手Marvin Gaye 1973年的《Let's Get It On》相似，陷入两起诉讼[3]。有趣的是，尽管中、美、日、韩等国设有著作权自愿登记制度，但多数国家遵循《伯尔尼公约》"自动保护"原则——作品诞生即获权，无需登记。为此，Ed Sheeran甚至在2023年纽约法庭弹唱歌曲自证原创性[4]，幸运的是，他唱服了陪审团！

Disputes over human-created works often revolve around "originality," frequently entangled with plagiarism claims. For instance, British singer Ed Sheeran’s Thinking Out Loud faced two lawsuits alleging similarities to Marvin Gaye’s 1973 Let’s Get It On [3]. Notably, while countries like China, the U.S., Japan, and Korea offer voluntary copyright registration, most jurisdictions follow the Berne Convention’s "automatic protection" principle—rights arise upon creation, without formalities. To prove originality, Ed Sheeran even performed live in a 2023 New York court [4], and luckily, his singing convinced the jury!

AI的登场让局面变得微妙。AI生成内容是否具备"独创性"？是否属于"人类智力成果"？这场辩论正酣。

AI’s emergence complicates this framework. Does AI-generated content possess "originality"? Is it a "human intellectual creation"? This debate is raging.

出于好奇，我向Suno输入指令："请创作一首女声城市民谣摇滚，致敬那些冲破困境的勇敢女孩"。顷刻间，一首名为《Rise Up Girl》[5]的歌曲翩然而至——歌词真挚、歌声优美、旋律流畅。欣喜之余，也不免惶惑：这般轻易诞生的佳作，当真属于我？

Out of curiosity, I asked Suno to "create a female vocal urban folk-rock song honouring the brave girls who have broken through." In a few seconds, a song called “Rise Up Girl” [5] arrived - heartfelt lyrics, beautiful vocals, and a smooth melody. While I was amazed, I also wondered: is this effortlessly born masterpiece really mine?

Suno回答[6]：付费用户拥有生成内容，免费用户仅限非商业使用，版权归属平台。据此，《Rise Up Girl》不属于我，因我未付分文。

Suno replied [6]: Paid users own generated content; free users are allowed to use those songs for non-commercial purpose, with ownership retained by Suno. As I did not pay, Rise Up Girl is not mine.

但著作权是否另有玄机？毕竟所有权与著作权未必同源。

But is copyright something else? After all, ownership and copyright may not be the same thing.

Suno解释：著作权问题很复杂，因地而异！纯AI生成音乐在美国不受保护，因法律仅承认人类创作，且输入指令不视为创作行为；而在其他地区或存可能，建议咨询当地机构。

Suno explained: Copyrights are complex, and vary by region! Pure AI-generated music would not qualify for copyright protection in the U.S. because the law only recognizes human creativity and writing the prompt does not constitute the creation of the song; while it might be possible in other regions, local authorities may be consulted.

美国版权局2025年1月发布的报告[7]明确指出：现行著作权体系仅保护含人类智力贡献的创作。当创作者以AI为工具，能实质性掌控表达元素，且成果彰显人类创造性主导时，方符保护要件。指南特别指出，当前技术阶段，仅通过文本指令（即便经过复杂优化）的创作模式，尚难满足人类实质掌控标准。

The report issued by the U.S. Copyright Office [7] in January 2025 makes it clear that the current copyright system only protects creations with human intellectual contribution. Protection applies when creators use AI as a tool to exert substantial control over expressive elements, with outputs reflecting human-led creativity. In particular, the Guidelines note that, at this stage of technology, a mode of creation based solely on text-based prompts (even with sophisticated optimization) does not yet meet the criterion of substantial human control.

2023年"Théâtre D'opéra Spatial案"是绕不过的例子。数字艺术家Jason Allen通过Midjourney输入624次指令、进行900次迭代优化（耗时80小时）完成的获奖画作，遭美国版权局不予登记，裁决[8]认为：文本指令仅为工具使用（生成方差32.5%证明不可控）；57.4%核心视觉元素由AI自主生成，人类未主导表达；成果属"机器衍生作品"，但独创性指令文本可单独受护。此案确立AI作品审查双重标准：人类对表达的主导控制与生成内容可预见性。这桩标志性案件，正静候联邦法院终裁。

The Théâtre D'opéra Spatial case of 2023 is a good example of this. Digital artist Jason Allen’s award-winning artwork, created via Midjourney through 624 prompts and 900 iterations (80 hours of work), was denied copyright registration by the US Copyright Office, which ruled [8] that prompts were mere tools (32.5% output variance proved unpredictability); 57.4% of core visual elements were autonomously generated by AI, with no dominant expression by human; and the output was a “machine-derived work”, although the creative text prompts may be protected by copyright separately. The case established dual criteria for AI works: human dominance over expression and output predictability. This landmark case awaits final federal court judgment.

美国版权局的报告另提及艺术家Kris Kashtanova的《Rose Enigma》案例。她通过Stable Diffusion将手绘草图转化为AI艺术作品，最终获批的著作权仅涵盖手绘创意内容。版权局在登记证书上特别备注"著作权限于可清晰识别且独立于AI生成内容的人类创作部分"。

The U.S. Copyright Office report also cites artist Kris Kashtanova’s Rose Enigma case. Her AI-generated artwork, created by Stable Diffusion from her hand-drawn sketch, received copyright only for the human-drawn portions, with a specific annotation by the U.S. Copyright Office on the registration certificate - " Registration limited to unaltered human pictorial authorship that is clearly perceptible in the deposit and separable from the non-human expression that is excluded from the claim."

依此标准，《Rise Up Girl》显然无法在美国获得著作权，我也绝非作者。Suno所言非虚。

By this standard, Rise Up Girl is clearly not copyrightable in the U.S., and I am in no way the author. What Suno says is true.

2023年11月，北京互联网法院对首例AI图片《春风送来了温柔》著作权案作出判决 [9]，认定未经许可使用AI生成图片构成侵权。法院认为，原告通过输入数百提示词并进行多轮调试，其过程凝结智力劳动，体现独创性，故AI生成图片受著作权法保护。该案裁判逻辑包括：1）创作控制——提示词复杂度与参数调整反映人类主导；2）工具属性——AI仅为技术手段；3）保护边界——独创性需个案判定，过于简略的指令恐难达标。此后多个类案判决亦循此理，显示中国司法的态度倾向于：只要满足"独创性"与"人类智力投入"，AI作品与传统作品无异，指令设计者即为作者。这一对AI创作"过程导向"的审查倾向，区别于美国版权局的"结果控制"标准。

In November 2023, the Beijing Internet Court issued a judgment on the first  copyright case relating to an AI image "Spring breeze brings tenderness" [9], holding that unauthorized use of the AI-generated images at issue constituted copyright infringement. The court reasoned that the plaintiff’s input of hundreds of prompts and iterative adjustments demonstrated intellectual labor and originality, thus qualifying the AI output for copyright protection. The court’s rationale included: 1) Creative control—prompt complexity and parameter adjustments reflect human dominance; 2) Tool nature—AI is merely a technical means; 3) Protection boundaries—originality requires case-by-case assessment, but overly simplistic prompts may fail. Several subsequent cases followed the same logic, signaling China’s judicial stance is inclined towards: AI works are equivalent to traditional works if they meet "originality" and "human intellectual labour" criteria—with AI as a tool and the prompter as author. This "process-oriented" approach contrasts with the U.S. "output control" standard.

该判决引发广泛讨论。支持者赞其平衡创新与权利保护。批评者忧AI作品易量产且含随机性，过度保护恐致权利滥用。更有观点认为，保护AI作品将扭曲著作权制度激励人类创作的初衷，因为AI主导内容生成过程，人类在其中的贡献比例过低。后者恰与美国版权局立场遥相呼应。

The judgment sparked widespread debate. Supporters praised it for balancing innovation and rights。 Critics warn of AI’s mass-production capacity and randomness, fearing overprotection could invite abuse of rights. Some even argue that protecting AI works will distort the original purpose of the copyright system to incentivize human creativity, as AIs dominate the content generation process and human contributions are disproportionately low. This echoes the U.S. Copyright Office’s position.

2025年4月18日，张家港市人民法院就一起AI生成儿童座椅著作权纠纷案作出的最新判决刚刚公布[10]。原告使用Midjourney输入指令“cute children’s chair with jelly texture, shaped like a butterfly…”（中译文：可爱儿童椅，果冻质感蝴蝶造型…）生成设计图后，将图片及指令发布在网上并寻求商业合作。被告借鉴类似提示词生成近似的AI设计并投产销售。法院判定涉案AI图片不构成作品且侵权不成立，理由是：1. 原告仅输入单次简单指令，既未通过多轮优化亦无人工修改（如Photoshop）体现个性化智力投入，且AI工具对相似提示词产出结果差异显著，证明人类缺乏对结果的实质控制；2. 原告提示词虽含材质、造型等参数，但属常规描述叠加且有在先类似设计元素，不构成独创性表达；3. 被告虽借鉴原告提示词，但其经AI重构及设计优化后已形成实质性差异风格。

On 18 April 2025, a most recent copyright dispute over AI-generated children’s chair designs ruled by the Zhangjiagang People’s Court was published [10]. The plaintiff used Midjourney with prompts like “cute children’s chair with jelly texture, shaped like a butterfly…” to create images, then shared prompts and images online, also seeking commercial partnerships. The defendant later used similar prompts to generate comparable designs and produced/sold products. The court rejected the infringement claim, stating: 1) The plaintiff’s minimal input—a single set of basic prompts without iterative refinements or manual edits (e.g., Photoshop)—demonstrated insufficient human control, evidenced by the AI’s inconsistent outputs from similar prompts; 2) The plaintiff’s prompts, though specifying elements like texture and shape, relied on common, unoriginal descriptors existing in prior designs; 3) The defendant’s final product, though inspired by the plaintiff prompts, diverged stylistically after AI regeneration and design refinements.

本案确立的裁判逻辑一定程度上吸收了美国版权局的审查标准，认为AI生成内容需体现人类对创作过程的实质性控制及结果的独创性，方受著作权保护，但又没有后者那么严苛。至于这是否预示中国司法对AI作品态度的转向，仍有待观察。

This decision has, to a certain extent, adopted the U.S. Copyright Office's criteria, holding that AI-generated content must embody substantial human control over the creative process and the originality of the output in order to be protectable by copyright, but is not as stringent as the latter. Whether this signals a shift in China's judicial attitude towards AI works remains to be seen.

我不禁在想，若原告先手绘蝴蝶状儿童椅的创意草图再结合手绘图及她输入的指令让AI生成设计图（类似于《Rose Enigma》案），本案判决结果会不同吗？

I can’t help but wonder: Had the plaintiff first hand- sketched the butterfly-shaped children's chair design and then combined hand-drawn draft with tailored prompts she inputted to have the AI generate the design (as in the Rose Enigma case), would the verdict have been different?

无论如何，我在Suno输入的寥寥数语，怕是无法让《Rise Up Girl》在中国成为著作权作品，我当然也主张不了作者身份。

Regardless, the few words I typed into Suno for Rise Up Girl would likely never secure copyright status in China, and I have no way of claiming authorship here.

根据英国同事Tia Ilana Acheampong的撰文[11]，英国早在1988年就已将“计算机生成的作品”纳入著作权保护对象，保护期50年，作者是“为创作该作品做出必要安排的人”。只需满足“必要安排”即可？这似乎只需要很低的人类参与度。虽然文章也指出，近期判例要求作品包含"作者自身智力创作"，门槛高于传统的"原创性"，但就法条的字面意思来看，我或许终于有机会（尽管心虚地）在英国主张《Rise Up Girl》的作者身份，并庆祝今年的世界知识产权日！

As early as 1988, according to an article by our UK colleague Tia Ilana Acheampong [11], “computer-generated works” were already copyrightable objects in the United Kingdom for a 50 year period of protection, with the author being “the person by whom the arrangements necessary for the creation of the work are undertaken”. Only the “necessary arrangements” need to be met that seems to require very little human involvement. While the article notes that recent case law require works to contain "the author’s own intellectual creation," a higher threshold than traditional "originality", still, the literal wording of the statute may finally give me the opportunity (albeit guiltily) to claim in the UK for the Rise Up Girl's authorship and celebrate World Intellectual Property Day this year!

最后，来听听《Rise Up Girl》吧！无论有没有著作权，这首人类思维火花与AI活力创造短暂互动创造的歌曲，真的很美！

Finally, take a listen to Rise Up Girl! Copyrighted or not, this song, created by the brief interaction between the spark of human thought and the energetic creation of AI, is truly beautiful!

"Rise Up Girl" / 《起来！姑娘》

(词、曲、声音均由Suno.com自动生成)

(Lyrics, Music & Vocal, all auto-generated by Suno.com)

She walked through the dark with trembling strides

她迈着颤抖的步伐，穿过黑暗

Shadows whispered lies she’d tried to hide

阴影低语着，她试图隐藏的谎言

The weight of the world pressed hard on her chest

世界的重量，重重压在她胸口

But deep down she knew she deserved her best

但内心深处，她明白要做到自己最佳

The city lights glared like truth in her face

城市的灯光，照着她真实的脸庞

Each step she took claimed her rightful space

迈出每一步，都走出她应有的空间

She shed every fear like leaves on stone

她放下所有恐惧，像扫去石头上的落叶

The girl she was had now fully grown

曾经的她，已经完全成长了

Rise up girl let your soul ignite

起来！姑娘，让你的灵魂点燃

Stand tall in the blaze of your own light

在自己的光芒中，屹立不倒

Break the chains that held you down so long

挣脱长久以来束缚你的枷锁

You’re the fire you’re the fight you’re the song

你就是火焰，你就是战斗，你就是赞歌

Each scar carved lines of wisdom and grit

每一道伤疤，都刻印着智慧和勇气

Every fall she faced taught her how to resist

每一次跌倒，都教会她如何反抗

Tall towers watched as she roared to the sky

高塔注视着，她向天空咆哮

No turning back no more living a lie

不再回头，不再活在谎言中

Through the street echoes she finds her voice

透过街道的回声，她找到了自己的声音

Every broken dream now feels like her choice

每一个曾破碎的梦想，如今都像她的主动选择

A mosaic of strength pieced from the pain

将痛苦的马赛克碎片，拼凑成力量之剑

She builds a world where she’ll reign

她建立了一个自己做主的世界

Don’t tell her she can’t don’t say she won’t

不要说她不能，不要说她不会

She’s the thundering past the trembling ghost

颤抖的幽灵已是过往，她已发出雷鸣

No silence can hush the storm she’s become

沉默无法平息她已转身变成的风暴

Her battle’s won her dawn has come

她的战斗已经胜利，她的黎明已经到来

参考 References:

[1] https://www.uspto.gov/sites/default/files/74629287.mp3

[2] https://news.cri.cn/2016-07-15/2c252d41-c002-385c-b896-ab6db811e3fb.html

[3] https://www.musicbusinessworldwide.com/ed-sheerans-second-thinking-out-loud-copyright-case-appealed-to-us-supreme-court/

[4] https://www.bbc.com/news/world-us-canada-65420860

[5] https://suno.com/song/54e6fcf1-99f5-4edc-b04e-10c1b1178034

[6] https://help.suno.com/en/articles/2746945

[7] https://copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf

[8] https://www.copyright.gov/rulings-filings/review-board/docs/Theatre-Dopera-Spatial.pdf

[9] http://www.ipforefront.com/m_article_show.asp?id=3697&BigClass=%E8%B5%84%E8%AE%AF

[10] https://mp.weixin.qq.com/s/CwFl3luPKqDbpO3iT8ScWw

[11] https://www.marks-clerk.com/insights/latest-insights/102k38x-who-owns-the-content-generated-by-ai/

--- 

联系作者 Contact the author

xhuang@marks-clerk.com.cn