Netflix uses AI in 300 titles, but human input still determines copyright.

In the second quarter of 2026, Netflix disclosed in its letter to shareholders that approximately 300 titles had used generative AI that year. Upon this announcement, many people’s first reaction was: If AI can now produce shows, who owns the copyright to these productions?

Netflix didn’t make any absolute claims. In its production guidelines for partners, it drew a clear line: any AI use involving the final cut, actor likenesses, personal data, or third-party content must undergo enhanced review and receive written approval before proceeding. In other words, scaling AI use is real—but copyright and liability aren’t automatically resolved just because AI is involved.

This question can be answered in one sentence: Copyright recognizes what a person did, not what the AI created. But this is just the first gate AI-generated films must pass. Netflix’s hundreds of films must clear five gates in total—we’ll go through them one by one.

Five gates

Let’s be clear: copyright is never granted to the “part created by AI,” but to “what the person did.”

The U.S. Copyright Office’s threshold is just four words: human authorship. If something is entirely generated by AI with minimal human involvement, it is excluded—no one can register it. But if a person genuinely contributes by selecting, arranging, or making creative modifications to the AI’s draft, the overall work can still qualify for copyright.

Of Netflix’s 300 titles, the vast majority use AI only in auxiliary stages such as post-production and visual effects. One example they disclosed: AI enhanced approximately 17 minutes of footage in a certain production, doubling the speed of production and halving the cost. Yet behind the final product, the work of writers, directors, editors, and actors remained unchanged. Human creativity carries substantial weight—AI is merely a tool, and copyright naturally belongs to humans.

The key criterion for the copyright office is not whether you used AI, but whether you can truly control what it produces. If you input a prompt and each generation yields something different—this kind of random, luck-based outcome is not recognized by law. There’s a classic analogy: if you scatter seeds in a garden, the resulting shapes are not under your control; the law won’t grant you ownership of an entire field of wildflowers just because you planted seeds. The same applies to AI: prompts are like sowing seeds; the specific output is “grown” by the AI, not truly “drawn” by you.

Of course, this analogy isn't perfectly precise. The difference is that if you extensively modify and refine the AI's output, reselecting and recombining elements, that portion becomes your own creation and is eligible for protection.

Two real registration cases best illustrate the threshold. One person used AI to generate a painting, entering prompts 624 times and applying post-processing software—but the Copyright Office still rejected it: no matter how many prompts you use, you’re only telling AI what you want; how the image is ultimately rendered is up to the AI, and you haven’t exercised “creative control.” Another person created an AI-assisted image, using image inpainting to revise it 35 times, expanding the canvas, adding elements, and submitting a 9-minute-58-second video documenting the creative process—the Copyright Office registered it. What was registered wasn’t the pixels generated by AI, but his selection, combination, and arrangement of those elements.

Having only prompts is usually not enough to make you an author.

The boundary between purely AI-generated and human-created content

This threshold is clearly outlined in the most authoritative precedents in the United States and the Copyright Office's reports.

Thaler v. Perlmutter, United States Court of Appeals for the District of Columbia Circuit, No. 23-5233 (March 18, 2025)

Human authorship is a bedrock requirement of copyright.

Human authorship is a fundamental requirement for copyright. (In this case, the applicant listed an AI system as the sole author of the work and was denied registration by the Copyright Office; however, the court explicitly clarified that this rule does not prohibit works created by humans using AI from receiving protection.)

U.S. Copyright Office Report on "Copyright and Artificial Intelligence, Part II: Copyrightability" (January 2025)

Generative AI output may be eligible for copyright protection only when a human author has established sufficient expressive elements: either the human-created portions are perceptible in the AI output, or the human has made creative arrangements or modifications to the output—but merely providing prompts is not enough. Incorporating AI-generated material into a larger human-created work does not prevent that work from obtaining copyright protection.

This gate regulates the writers. The 2023 Hollywood writers' strike, which lasted 148 days, resulted in a contract (WGA 2023 MBA) that firmly defined several aspects of the relationship between AI and writers.

AI is not a screenwriter. Content generated by AI does not qualify as “literary material” or “source material” for determining authorship. This means that even if a company provides a screenplay draft generated by AI to a screenwriter, the screenwriter remains the “first author” of the project, and their credit and compensation are unaffected. The company also has an obligation: if any AI-generated content is included in the materials provided to the screenwriter, it must be disclosed truthfully. Conversely, the company cannot compel the screenwriter to use AI.

Artificial Intelligence Provisions of the 2023 Writers Guild of America Contract (WGA 2023 MBA)

Neither traditional AI nor generative AI is a writer, so no written material produced by traditional AI or GAI can be considered literary material.

Neither traditional AI nor generative AI is a screenwriter, so written content produced by AI cannot be considered "literary material."

Same as above (attribution and compensation)

When AI-generated material provided by the company has not been previously published, "such material does not constitute assigned material (for calculating compensation) or source material (for determining attribution)," and the writer is therefore regarded as the first author of the project.

Same as above (disclosure obligations and mandatory use)

The Company must disclose to the writer if any materials provided to the writer have been generated by AI or incorporate AI-generated content.

…the company can't require the writer to use AI software (e.g., ChatGPT) when performing writing services.

Companies must disclose to writers whether the materials provided to them are generated by AI or contain AI-generated content; companies cannot require writers to use AI software (e.g., ChatGPT) when performing writing services.

These rules apply to Hollywood, but their logic holds for any company using AI to generate content: AI can help you draft, but it doesn’t take credit, nor does it change who should be paid. If your company uses AI to write proposals or copy, it’s best to clarify upfront who the “author” is and who bears responsibility.

This gate controls the actor's face and voice.

Today, AI can scan an actor’s facial features, voice, and performance traits and store them as data. Later, their performance can be “generated” without their physical presence. This data is called a digital twin.

The same contract negotiated during that strike established three safeguards by SAG-AFTRA for digital replicas: producers must notify actors at least 48 hours in advance before creating a digital replica; they must obtain the actor’s “clear and conspicuous” consent, which cannot be buried in standard terms but must be signed or stamped separately; and they must compensate the actor. When used, producers must also provide the actor with a “reasonable and specific description” detailing how and where the replica will be used. For any new project or different use, fresh consent must be obtained.

Netflix’s own rules also align with this: altering an actor’s age, synthesizing voices, or substantially changing the emotion and intent of a performance are all listed as requiring written consent.

SAG-AFTRA 2023 Motion Picture / Television Contract (Digital Replica Provisions)

48-hour advance notice prior to services for creation… Consent required… Must be clear and conspicuous… Separately signed or initialed by the performer… Must include a reasonably specific description of the intended use.

A 48-hour notice is required prior to creating a digital replica; consent must be obtained; consent must be clear and conspicuous, signed or initialed separately by the individual; and must be accompanied by a reasonable and specific description of the intended use.

The actor team fears "blanket authorization" most. A contract that vaguely states "authorization to use the likeness" without limiting duration, context, or number of uses is equivalent to permanently selling one's face and voice—no different from signing a contract of indentured servitude. The prudent approach is to grant authorization on a per-project basis, authorizing only one specific use at a time.

This also has a precedent in China. The Beijing Internet Court ruled on a case involving AI-generated voice used for product promotion, determining that if an AI-synthesized voice can be recognized by an average person as belonging to a specific individual, it falls within the scope of protection for voice rights, and unauthorized use constitutes liability. The standard for judgment is "identifiability": if it’s close enough to be recognized as you, then it counts as using you.

This gate raises a more upstream question: Whose data was used to train this AI?

One of the红线 Netflix draws for its partners is: do not use materials from artists, actors, or other rights holders to train or fine-tune models without proper legal authorization. It also requires partners to review the terms of the tools they use to ensure the AI company does not retain or reuse inputs for further training.

Why the strict oversight? Because issues at the training stage propagate all the way to the final product. If an AI tool trains on unauthorized content, the images, music, and other elements it generates may constitute “substantial similarity” to the original works. When such a finished product is publicly released, the producer may face not only infringement claims from the original rights holders—but in severe cases, the entire project’s core copyright could be invalidated due to unclear ownership, forcing it to be taken down and rendering all prior investments worthless. This liability flows back and forth among the producer, the AI provider, and the deliverer through warranty and indemnity clauses in their contracts.

Netflix Using Generative AI in Content Creation (Partner Guidelines, Training Data Section)

Do not train or fine-tune models using material from artists, performers, or other rights holders unless you have the proper legal clearance.

Material from artists, performers, or other rights holders must not be used to train or fine-tune models unless appropriate legal authorization has been obtained.

This gate was set up by Netflix itself. It lists five low-risk principles for partners; all five must be met to use AI without additional approval.

Netflix Generative AI in Content Creation (Five Low-Risk Principles)

• The output will not copy or substantially reproduce identifiable features of materials owned or protected by copyright of others;

• The generation tool used does not store, reuse, or use your input or output data for training;

• Designed for enterprise-grade security environments to protect input content;

• The generated material is temporary and does not become part of the final deliverable;

• Do not use generative AI to replace or generate performances by actors or work covered by unions without obtaining consent.

If any one question cannot be answered with a “yes,” the review must be escalated and written approval required. This gate is not imposed by law, but rather a safeguard the platform has put in place through contracts with itself and its partners. It fills the gaps left by the previous layers: the details that laws and unions cannot cover, now addressed by the platform’s own rules and approval processes.

Putting these gates together makes one thing clear: AI film and video are not governed by a single law or agency. The Copyright Office decides whether a work is valid, the Writers Guild regulates attribution and compensation, the Actors Guild oversees faces and voices, the cleanliness of training data is enforced layer by layer through copyright law and contracts, and platforms add their own approval process. None can replace the others.

This highlights the most practical judgment for AI content companies: a simple statement from a model provider saying “our data is compliant” cannot replace your own thorough rights clearance process. A vague written promise is far less valuable than a documented, traceable process. In the event of a dispute, judges don’t care who you believed—they care what evidence you left behind.

General audience: Whether AI was used in the movie you watch on Netflix is often undetectable and rarely affects your viewing experience. What truly matters are those “synthetic performances”—when a familiar actor appears in scenes they never actually filmed—whether there was clear and explicit consent behind it is the dividing line between legal and illegal use of that imagery.

Writers and actors: Remember two terms. Writers remember: "AI is not an author"—your credit and compensation should not be reduced just because the company used AI. Actors remember: "One play, one license"—do not sign agreements without time limits, specific scenes, or profit-sharing terms.

AI content entrepreneurs: This is the group with the most critical stakes. If you're working on AI films, AI short dramas, or providing tools and post-production services for such projects, your contract must clearly break down at least seven sets of clauses: authorization for training data, no retention of inputs or outputs, documented records of human creative contributions, individual consent from actors, disclosure of AI usage, guarantees of rights to the final deliverables, and clear allocation of liability for infringement. All seven sets are essential—none can be omitted.

Netflix’s hundreds of AI-generated works can claim copyright not because AI has become more powerful, but because humans are still involved. Copyright recognizes human creation, and unions protect human livelihoods. The platform’s approval gates safeguard its own risks. While AI has lowered production costs, all five of these checkpoints remain intact—and due to scaling, they are now all exposed on the same production line for the first time.

Those looking to profit on this line should treat every gate as real, rather than betting that rules will loosen. Taking precautions upfront is far cheaper than hiring a lawyer after something goes wrong.

If you're working on an AI content-related project and need someone to review your contracts, feel free to reach out.

•Netflix Q2 2026 Shareholder Letter (SEC Filing Exhibit 99.1)

•Variety: About 300 Netflix Programs Have Used Generative AI This Year (2026)

•IndieWire: Netflix Co-CEO Explains How Gen-AI Was Used in 300 Different Titles

• Netflix partner production guidelines: "Using Generative AI in Content Production" (Netflix Partner Help)

• U.S. Copyright Office Report on "Copyright and Artificial Intelligence, Part II: Copyrightability" (January 2025)

• U.S. Copyright Office Guidelines on Copyright Registration of Works Containing Artificial Intelligence-Generated Material (37 CFR Part 202)

•Thaler v. Perlmutter, United States Court of Appeals for the District of Columbia Circuit, No. 23-5233 (130 F.4th 1039, March 18, 2025)

• AI provisions of the 2023 Basic Agreement of the Writers Guild of America (WGA 2023 MBA)

• 2023 SAG-AFTRA Motion Picture and Television Contract (Digital Replicas Provision)

Author: Lü Yinghui