OpenAI Sora: The Legal and Copyright Lessons from AI Video Generation

OpenAI’s Sora — the text-to-video AI model first announced in February 2024 and rolled out through Sora 2 in late 2025 — has become one of the most legally consequential AI products of the current generation. Within weeks of broader release, Sora generated thousands of viral videos that simultaneously demonstrated the technology’s remarkable capabilities and exposed the fundamental tensions between AI video generation and existing copyright, personality rights, and consent frameworks.

The controversies around Sora — including deepfake videos of identifiable persons (both living and deceased), generation of content closely mimicking copyrighted characters and intellectual property (Disney, Pixar, Studio Ghibli, sports brands), and questions about training data and rights clearance — make the product a leading case study for the application of the EU AI Act (Regulation 2024/1689), Italian Law 132/2025, and traditional copyright and personality rights frameworks to AI video.

This guide analyses the legal questions raised by Sora and similar AI video products, with attention to the Italian framework. For the broader copyright framework, see our master pillar guide to copyright law in Italy and Europe. For AI photography and the Eldagsen case, see our AI photography guide. For moral rights and AI deepfake in cinema, see our moral rights in film guide.

The Sora controversy

OpenAI’s Sora generates photorealistic and stylized video from text prompts, with capabilities far exceeding previous generations of AI video tools. When opened to broader public use, the product immediately generated three categories of legally problematic content:

• Deepfake videos of identifiable persons: AI-generated footage of celebrities, public figures, and private individuals in situations they had not consented to depict — including deceased celebrities (Michael Jackson, Robin Williams, Whitney Houston) reanimated in new “performances”;
• Videos closely mimicking copyrighted characters and brands: Sora-generated content featuring characters and visual aesthetics from Disney, Pixar, Studio Ghibli, Marvel, Nintendo, and other rightsholders’ protected works;
• Videos in the style of identifiable creators: AI imitations of distinctive directorial styles, animation studios’ visual signatures, and individual artists’ aesthetic approaches.

The response from rights holders was rapid. Letters from major Hollywood studios and animation companies sought immediate restrictions. Estates of deceased celebrities demanded removal of unauthorized AI-generated content. Multiple lawsuits were filed in US and international courts. OpenAI announced progressive restrictions — opt-in/opt-out frameworks for likenesses, character recognition filters, training data transparency commitments — but these measures faced criticism both for being insufficient (from rights holders) and for chilling legitimate uses (from creators).

Deepfake and personality rights

Under Italian law, AI-generated video depicting identifiable persons engages multiple frameworks simultaneously:

• Article 10 of the Italian Civil Code: prohibits publication of another person’s image where it harms their decorum or reputation, or is otherwise displayed without authorization;
• Articles 96-97 of the Italian Copyright Act: a person’s portrait cannot be exhibited, reproduced, or commercialized without consent (with limited exceptions for notorious persons in their public role);
• GDPR (Regulation EU 2016/679): a person’s image is personal data, requiring valid legal basis for processing;
• Italian Law 132/2025 on artificial intelligence: specific transparency obligations on AI-generated content depicting persons.

For deceased persons, Italian personality rights framework permits heirs to enforce posthumous protections against unauthorized AI use of the deceased’s image — an increasingly important enforcement track as AI products generate content using deceased celebrities’ likenesses. See our moral rights of heirs guide for analysis of heir-controlled estates and modern AI applications.

For Italian celebrities like Sophia Loren, Marcello Mastroianni, or Federico Fellini (deceased), AI generation of new “performances” or imagery using their likenesses without heir consent would face simultaneous personality rights and moral rights claims, with substantial damages potential.

Intellectual property infringement

The IP questions raised by Sora’s output divide into two related but distinct categories:

Direct character reproduction

Where Sora generates content depicting specific copyrighted characters (a Mickey Mouse in a new scenario, a Pixar-style animated character closely resembling specific Pixar IP), the output reproduces copyrighted expression. The legal analysis under Italian law is straightforward: reproduction of copyrighted characters requires authorization, and AI generation does not transform the analysis. The producer of the resulting content (the user prompting Sora, OpenAI’s complicity in providing the tool, or both) faces copyright infringement liability.

Style imitation

More legally complex: Sora generates content “in the style of” identifiable studios, directors, or artists without directly reproducing specific copyrighted works. Under traditional Italian copyright analysis, style itself is not protected by copyright — the idea/expression dichotomy excludes general aesthetic approaches from protection (see our idea/expression dichotomy guide).

However, the Sora situation reveals important nuances:

• Where style imitation is so close that it amounts to reproduction of specific protected expressions, copyright applies;
• Where the imitation creates consumer confusion about source or association, trademark and unfair competition claims arise;
• Where the imitation is sufficiently specific to identify the artist whose style is mimicked, moral rights (paternity through false attribution) and personality rights may apply;
• Where commercial value flows from the style association, unjust enrichment claims arise.

AI training data questions

Sora’s capabilities depend on training on massive datasets including, almost certainly, copyrighted videos, films, animations, and other audiovisual content. Under the EU framework:

• DSM Directive Articles 3 and 4: text and data mining (TDM) is permitted for scientific research (Article 3, no opt-out) and for any purpose including commercial AI training (Article 4, with rights holder opt-out);
• Italian implementation: Articles 70-ter and 70-quater LDA transpose the framework;
• Opt-out mechanisms: rights holders must signal opt-out in machine-readable manner (robots.txt, metadata, platform settings, terms of use).

For video content, the opt-out infrastructure is much less developed than for text or images. Many video rights holders had not implemented meaningful opt-outs before Sora’s training was conducted. The retroactive application of opt-out frameworks remains legally contested, with multiple international disputes pending.

EU AI Act and Italian Law 132/2025

The EU AI Act (Regulation 2024/1689) applies directly to Sora and similar foundation model providers, with specific obligations:

• Foundation model obligations: providers must publish training data summaries, implement risk management, comply with copyright opt-out frameworks;
• Synthetic content labeling: AI-generated content distributed commercially must be identified as such;
• Deepfake transparency: AI-generated content imitating identifiable persons faces specific disclosure obligations;
• Risk classification: certain AI uses (biometric identification, automated content moderation) are subject to additional regulatory scrutiny.

Italian Law no. 132 of 22 September 2025 supplements the EU framework with national-specific provisions, including:

• Identification requirements for synthetic content distributed in Italy;
• Specific personality rights protections against unauthorized AI use of likenesses;
• Liability framework for AI-generated content infringing copyright, personality rights, or other protected interests;
• Sectoral applications in media, journalism, and audiovisual production.

For Italian audiovisual productions accessing the cinema tax credit, the mandatory AI clause under Article 7 §6 of D.I. MiC-MEF 225/2024 requires specific contractual provisions on AI use, consent, and transparency — see our tax credit guide.

Lessons for producers and platforms

The Sora experience offers concrete lessons for content producers, AI platforms, and creators:

For traditional producers and creators

• Implement opt-out: technical opt-out (robots.txt, metadata) and contractual opt-out (terms of use) for all content portfolios;
• Document creative provenance: maintain records distinguishing human creative work from AI-assisted or AI-generated content;
• Monitor for derivative works: automated monitoring for AI-generated content closely derivative of your works;
• Update contracts: standard contracts (talent, music, photography) should specifically address AI training and likeness use;
• Estate planning for moral rights: for talent, structuring estates to ensure heir enforcement against unauthorized AI use posthumously.

For AI platforms and AI-using businesses

• Training data compliance: respect opt-outs, document training data provenance, prepare for AI Act transparency requirements;
• Output filtering: implement detection mechanisms for copyrighted characters, identifiable persons, and protected IP;
• Consent frameworks: opt-in mechanisms for persons whose likeness can be generated, with clear scope limitations;
• Disclosure: clear synthetic content labeling per AI Act and national requirements;
• Tax credit compliance: for Italian-funded productions, AI clause compliance under D.I. 225/2024.

How DANDI supports AI-related clients

DANDI.media advises producers, talent, creators, AI platforms, and rights holders on AI-related legal questions:

• AI opt-out frameworks for content portfolios;
• Production contracts with AI use provisions and tax credit AI clause compliance;
• Personality rights enforcement against unauthorized AI deepfakes;
• Moral rights claims for AI use of deceased authors and performers;
• EU AI Act and Italian Law 132/2025 compliance for AI businesses;
• Cross-border AI litigation coordination.

For consultation, book directly with Avv. Claudia Roggero or Avv. Donato Di Pelino.

Related guides

Dandi Law Firm provides legal assistance in several Practice Areas. Check out our Services or contact Us!