21世纪经济报道记者王俊 肖潇 报道 站在2026年的初始回顾2025年，似乎契合了那句"我们总是通过后视镜驶向未来。" 2025年春节，DeepSeek出圈，不仅撕开了硅谷铸造的AI铁幕，也终结了前两年AI技术与商业的晦暗不 明。自此AI一路高歌、产品迭代密集，年末的"王炸"豆包手机智能体，更是直接把AI时代的入口之争摆 上了台面，新的利益格局正在塑形。 与此同时，传统地盘的合规争议进一步深化。内容平台上，张兰、汪小菲、张雪峰等账号被封禁，"黑 红也是红"的流量逻辑开始失效；外卖战场里，阿里、美团、京东短兵相接，在核心地盘上攻防，"烧钱 补贴"玩法再被治理。 回顾2024年的时候，我们曾用"新旧之间"来贯穿主线。到了2025年，张力不但没有缓解，反而在更多领 域交织，演变为更复杂的生态博弈。 混沌的格局中，有一个颠扑不破的真理：抢人、抢用户。而随着AI产品向更广泛人群扩散，纠纷逐渐 浮出水面。过去一年里，AI谣言更多了、真假更难辨了， AI情感陪伴带来的问题更频繁了，AI黑灰产 开始渗透进日常生活了......."用户究竟会使用AI做什么"，直接牵动大模型公司的商业策略与合规设计， 已经成为不得不回答的问题 ...

Core Viewpoint - The importance of safety in AI products is emphasized, as any lapse could lead to significant consequences for the industry and public trust [1][3]. Group 1: AI Safety Concerns - The security testing results for AI hardware products are concerning, with vulnerabilities that could allow for malicious activities, such as using AI glasses for covert surveillance [3][5]. - The current state of security for embodied intelligent robots is described as "full of holes," indicating multiple vulnerabilities that could lead to harmful actions [3][5]. - There is a significant gap in understanding the value of security among many emerging AI startups, with some companies neglecting to establish dedicated security teams until late in their development [5]. Group 2: AI Mobile Assistants - The launch of the AI mobile assistant "Doubao" has sparked controversy regarding its system-level permissions and potential risks to user privacy [6][7]. - The regulatory landscape for AI assistants is evolving, with calls for balanced oversight that prevents both excessive regulation and unchecked growth [7][8]. Group 3: AI Content Regulation - New regulations require AI-generated content to have visible and hidden identifiers to ensure accountability and traceability [9][10]. - There are emerging black market activities focused on removing AI identifiers, posing challenges for content regulation [10][12]. Group 4: AI and Youth Protection - The rise of AI companionship products has raised ethical concerns, particularly regarding their impact on youth, with incidents of self-harm linked to AI interactions [13][14]. - Companies are implementing measures to protect minors, such as usage restrictions and behavior analysis to identify underage users [14]. Group 5: AI Copyright Issues - Ongoing disputes regarding AI copyright focus on the legality of using copyrighted material for training data and the responsibilities of AI platforms [15][17]. - Recent statistics indicate a significant number of copyright infringement cases related to AI in the U.S., highlighting the urgency of resolving these issues [17]. - Collaborative agreements between AI companies and copyright holders, such as the partnership between Disney and OpenAI, are seen as potential pathways to resolve disputes and foster innovation [19].

通过持有OpenAI的股权，迪士尼获得了一张进入AI核心圈的门票。作为OpenAI的股东之一，迪士尼自 然获得了技术优先权，将广泛使用ChatGPT和其他AI工具在内容和产品开发等场景中。这项合作让迪士 尼能够进入社交平台快速增长的用户生成短视频领域。迪士尼CEO艾格在交易公布后对媒体称，与 OpenAI达成的授权合作旨在"为迪士尼打开新的数字娱乐空间并吸引更多年轻用户"。市场对这一交易 的回应是积极的，迪士尼股价迎来了一波持续上涨。 迪士尼刚宣布给OpenAI砸下10亿美金搞'世纪联姻'，反手就逼着谷歌把AI生成的米老鼠、死侍全部下 架！一边当最大的金主，一边举最狠的屠刀，本期视频给大家从资本、版权、行业生态几个角度详细拆 解这位全球娱乐巨头到底在下什么大棋？ 近日，迪士尼宣布把自家200多个角色IP开放给Sora做视频生成，同时宣布对OpenAI投资10亿美元，双 方正式达成深度战略合作伙伴关系及内容授权协议。几乎在同一时间，迪士尼法务部向谷歌发出了版权 警告，要求其旗下YouTube及相关平台立即下架利用AI生成的，包含米老鼠、死侍、钢铁侠等经典IP形 象的视频内容。谷歌迅速响应，数小时内下架了相关违规 ...

Core Viewpoint - Disney has issued a cease-and-desist letter to Google, accusing the company of large-scale copyright infringement through the use of AI models and related services, which generate and distribute images and videos involving Disney's classic intellectual properties for commercial use [1][3]. Group 1: Legal Actions and Allegations - Disney has engaged Jenner & Block law firm to send a letter to Google's General Counsel, stating that Google has unauthorizedly copied a significant number of Disney works as training material for its AI, resulting in copyright infringement [3]. - The letter highlights that the AI-generated infringing content includes characters from popular animated films such as "Frozen," "The Lion King," "Moana," and "The Little Mermaid," as well as characters from blockbuster franchises like "Deadpool," "Guardians of the Galaxy," and "Star Wars" [3]. - Disney has provided an example of an image of Darth Vader generated by Google's AI application through text prompts to illustrate the infringement [3]. Group 2: Demands and Industry Context - Disney demands that Google immediately cease the copying, displaying, distributing, or creating derivative content based on Disney characters across various platforms, including YouTube mobile, YouTube Shorts, and other Google AI output products [3]. - The company insists that Google implement technical restrictions in its AI services and all products integrating such AI functions to prevent future infringement at the source [3]. - This action by Disney aligns with its previous cease-and-desist letters to Meta and Character.AI, and it echoes a joint lawsuit with NBCUniversal and Warner Bros. Discovery against Midjourney and Minimax, demonstrating Disney's firm stance on protecting its intellectual property rights [3][4].

Core Viewpoint - The article discusses a landmark case in China regarding AI copyright infringement, highlighting the responsibilities of both users and AI platforms in the context of copyright law [1][4][12]. Group 1: Case Details - The Shanghai Jinshan District People's Court ruled on November 3 that a user infringed copyright by using images of the character "Medusa" from the anime series "Dou Po Cang Qiong" to train an AI model, resulting in a compensation of 50,000 yuan [1][4]. - The AI platform involved was not held liable as it had promptly removed the infringing model and updated its keyword filters after receiving a lawsuit, fulfilling its "notice-and-takedown" obligations [1][12]. - The user, identified as Li, used over 20 images of "Medusa" to create a model that allowed others to generate similar images, which the court deemed a violation of the original copyright holder's rights [4][12]. Group 2: Implications for AI Platforms - The court's decision sets a precedent for how AI platforms are treated under copyright law, emphasizing the need for platforms to respond quickly to infringement complaints and implement effective monitoring systems [1][12][14]. - The ruling aligns with previous cases, such as the "Ultraman AI infringement case," where courts found that platforms are not directly liable if they do not participate in the infringement and take appropriate actions upon notification [12][13]. - Legal experts suggest that AI companies should enhance their complaint handling processes, improve their content review systems, and clearly inform users about copyright risks when using training features [14]. Group 3: Industry Context - The popularity of "Dou Po Cang Qiong" has led to widespread AI-generated content, with many users creating videos and images based on the series, raising questions about the balance between creative expression and copyright protection [5][12]. - The emergence of user-generated content (UGC) communities and AI model fine-tuning presents new challenges for copyright enforcement, necessitating a reevaluation of the responsibilities of AI platforms [13][14].

Core Viewpoint - The case marks a significant development in the ongoing AI copyright dispute, establishing a precedent for how AI platforms and users may be held accountable for copyright infringement in China [1][5]. Group 1: Case Details - The Shanghai court ruled that the user, who used images of the character "Medusa" from the anime series "Dou Po Cang Qiong" to fine-tune a large model, infringed copyright and was ordered to pay 50,000 yuan in damages [1][3]. - The AI company was found not liable for the infringement as it had promptly removed the infringing model and updated its keyword filters upon receiving the lawsuit [1][5]. - The user’s actions were deemed to meet the standards of "access" and "substantial similarity," thus violating the original copyright holder's rights [3][4]. Group 2: Implications for AI Platforms - The ruling emphasizes the need for AI platforms to have effective complaint mechanisms and to act promptly on infringement claims to avoid liability [5][7]. - The distinction between AI platforms as "content providers" versus "technology providers" is crucial, with the court suggesting that platforms must be aware of potential infringement risks [6][7]. - Legal experts suggest that AI companies should enhance their compliance measures, including improving user agreements and providing clear warnings about copyright risks when using training functions [7].

Core Viewpoint - The article discusses the emergence of Sora 2 by OpenAI, highlighting its potential to revolutionize the AI video generation landscape and the competitive dynamics it introduces in both domestic and international markets [2][5][19]. Group 1: Sora 2's Features and Innovations - Sora 2 has made significant advancements over its predecessor, including the first-time synchronization of audio and visuals, improved physical accuracy, and enhanced resolution and detail, marking a pivotal moment in AI video generation [7]. - Compared to industry averages, Sora 2 excels in key performance metrics such as physical consistency, multi-shot storytelling, and audio-visual synchronization, outperforming by over 40% [7]. - The introduction of the "Cameo" feature allows users to create digital avatars and authorize others to use them, raising concerns about potential copyright infringement and misuse of digital assets [8][12]. Group 2: Market Dynamics and Challenges - Despite the excitement surrounding AI video models, the commercial viability remains uncertain, with high costs and unclear revenue prospects being significant barriers [5][12]. - OpenAI's new copyright policy aims to give IP owners more control over how their characters are used, but the challenge lies in balancing pricing and preventing IP misuse [12][14]. - The article suggests that while Sora 2 may resemble an AI version of social media platforms, it faces significant hurdles in achieving widespread user acceptance and creating sustainable IP assets [16][17]. Group 3: Competitive Landscape - The launch of Sora 2 has led to market reactions, including a drop in Meta's stock price, reflecting concerns about the potential disruption to existing social media ecosystems [16]. - The article emphasizes that while AI tools can democratize content creation, the true differentiator remains the creativity and storytelling ability of individual users, which AI cannot replicate [16][17]. - The current state of AI-generated content is characterized by high homogeneity and questionable quality, raising concerns about the future of artistic skills and content diversity [17].

Core Viewpoint - The article discusses the ethical and copyright issues surrounding AI-generated videos of deceased celebrities, particularly focusing on the case of Robin Williams and the implications of OpenAI's Sora 2.0 release, which has sparked significant controversy and backlash from family members and industry stakeholders [1][2][3]. Group 1: AI Video Generation and Controversy - The release of Sora 2.0 has led to a surge in AI-generated videos featuring Robin Williams, raising concerns about the manipulation of his image and voice without consent [1][3][5]. - Robin Williams' daughter has publicly condemned the creation of AI videos of her father, emphasizing the emotional distress it causes to the family and the disrespect it shows to his legacy [5][6]. - The rapid adoption of Sora 2.0, which reportedly surpassed one million downloads within five days, highlights the growing demand for AI-generated content, but also the challenges of regulating its use [5][6]. Group 2: Legal and Ethical Implications - The article outlines the legal framework in China regarding the posthumous rights of deceased individuals, indicating that family members can claim rights over the deceased's image and voice, which complicates the use of AI in recreating these figures [8][9]. - OpenAI has faced pressure from various stakeholders, including Hollywood unions and family members, to establish clearer boundaries regarding the use of deceased individuals' likenesses in AI-generated content [13][14]. - OpenAI has adjusted its copyright policy from an opt-out to an opt-in mechanism, allowing public figures to control the use of their likenesses in Sora-generated videos, although this does not address the rights of deceased individuals [14][15]. Group 3: Industry Response and Future Directions - The article notes that the backlash against AI-generated content is not isolated, as other companies in the industry have faced similar legal challenges and public outcry regarding copyright infringement [13][16]. - There is a call for a more structured approach to the ethical use of AI in recreating public figures, with suggestions for obtaining explicit consent from deceased individuals' estates and establishing clearer guidelines for AI platforms [9][16]. - The ongoing debate highlights the tension between artistic expression and the rights of individuals, suggesting that the industry is still in the process of finding a balance between innovation and ethical responsibility [16].

Core Insights - OpenAI launched its most powerful video generation model, Sora 2.0, which achieved over one million downloads within five days, surpassing the initial speed of ChatGPT [1] - The rapid adoption of Sora 2.0 has reignited long-standing concerns regarding AI copyright issues, particularly as users began generating fan videos using well-known intellectual properties (IPs) [2][3] - In response to the backlash, major Hollywood agencies and companies like Disney are pressuring OpenAI to take responsibility for copyright infringement, leading to a strategic shift in Sora's operational policies [3][4] Legal Context - The controversy surrounding Sora 2.0 stems from the "opt-out" mechanism that allowed the generation of copyrighted content unless explicitly requested to be removed by the copyright holders, which has been criticized for potentially leading to systemic infringement [4][8] - OpenAI's new "opt-in" policy, announced by CEO Sam Altman, aims to establish a revenue-sharing model with copyright holders, marking a significant shift in the relationship between AI companies and IP owners [4][21] - The legal challenges faced by AI companies include the legitimacy of using copyrighted works for training AI models and the risk of generating content that closely resembles existing copyrighted works [9][10][13] Business Model Implications - The new revenue-sharing model proposed by OpenAI seeks to redefine user-generated content as interactive fan creations, providing copyright holders with more control over their IPs and potential revenue streams [18][19] - This model is compared to YouTube's copyright revenue-sharing system, which could incentivize more creative content while offering copyright holders new monetization opportunities [19][22] - However, the implementation of this model faces challenges, including the complexity of tracking and attributing copyright elements in generated content, as well as the need for a clear and fair pricing structure for IP licensing [20][23] Industry Outlook - The shift from litigation to collaboration between AI companies and copyright holders reflects a broader trend in the industry, where the focus is on finding mutually beneficial solutions to copyright disputes [5][21] - The ongoing debate over AI-generated content and copyright distribution highlights the need for updated legal frameworks and standards to address the unique challenges posed by generative AI technologies [22][23] - OpenAI's approach signals a potential transition for the AI industry from unregulated growth to a more structured licensing phase, emphasizing the importance of innovative institutional designs to navigate the complexities of copyright in the AI era [23]

Core Viewpoint - The lawsuit filed by major Hollywood studios against Chinese AI company MiniMax highlights the ongoing conflict between technology and copyright, as MiniMax is accused of creating a "piracy business model" by using protected characters to train its AI system and generate unauthorized content [1][2]. Group 1: Lawsuit Details - The lawsuit was filed on September 16, 2025, by Disney, Universal Pictures, and Warner Bros, accusing MiniMax of systematically copying valuable copyrighted characters to profit from unauthorized videos [1][2]. - MiniMax's users can generate high-quality videos featuring iconic characters like Spider-Man and Darth Vader by simply inputting prompts, which has raised significant concerns among the studios [2]. - The studios had previously sent cease-and-desist letters to MiniMax, but the company did not respond substantively, prompting the legal action [2]. Group 2: MiniMax's Global Expansion - Prior to the lawsuit, MiniMax had rapidly expanded internationally, claiming over 157 million users across more than 200 countries [3]. - The company launched several AI applications, including Glow and Talkie, with Talkie achieving over 11 million monthly active users by 2024, primarily from the U.S. market [3]. - MiniMax's revenue was projected to exceed $70 million in 2024, with Talkie being a significant contributor to this growth [3]. Group 3: Financing and Market Position - MiniMax has been a capital darling, nearing completion of a $300 million funding round in 2025, which would elevate its valuation to over $4 billion [4]. - The company previously raised $600 million in March 2024, with Alibaba leading the round, and had a valuation of $2.5 billion at that time [4]. Group 4: Implications of the Lawsuit - The lawsuit poses a significant threat to MiniMax's financing and IPO plans, with potential damages reaching hundreds of millions of dollars, which would be a substantial burden for a company with annual revenue of around $70 million [5][6]. - The plaintiffs are seeking either profits gained from the infringement or statutory damages of up to $150,000 per infringed work under U.S. copyright law [5]. Group 5: Future Outlook - Legal experts suggest that the outcome of the case may hinge on the ongoing debate regarding whether AI training constitutes fair use, which remains a contentious issue in U.S. courts [7]. - A recent settlement involving Anthropic, which paid $1.5 billion to resolve a similar lawsuit, may serve as a reference for MiniMax, although the company faces unique challenges as a rapidly expanding Chinese AI firm in the U.S. market [8][9]. - The case could have broader implications beyond legal matters, reflecting the complexities of U.S.-China tech competition and the evolving landscape of copyright in the age of AI [9].