Learning objectives

• O1 — Classify a film's human-authored and AI-generated elements, and distinguish copyright protection from contractual permission and infringement clearance.
• O2 — Evaluate an AI litigation claim by separating allegations, procedural outcomes, fair-use arguments and dated tracker statistics.
• O3 — Identify the consent, compensation and jurisdiction-specific identity protections that apply to a recognizable replica or a synthetic performer.
• O4 — Apply the distinct publication, platform-removal and child-protection rules to harmful synthetic media without overstating constitutional rulings.
• O5 — Assemble a U.S./EU release checklist that combines provenance, disclosure, platform requirements and evidence of human contribution.

The production turning point

AI & Media: The Legal Landscape. Educational information, not legal advice. This reference explains the legal issues behind the intro video. It focuses on U.S. law, covered SAG-AFTRA agreements and selected European/U.K. rules, using a research cut-off of 13 November 2026. Laws, contracts, litigation and distribution policies can change; a particular release may need advice in its relevant jurisdictions.

Generative AI changes the scale and location of creative choices. A newsroom can draft summaries, translate interviews or generate an anchor voice. A film team can previsualize shots, build backgrounds and synthesize performances. Tasks once split among writers, editors, actors and postproduction vendors may now pass through one generator. Faster production makes it easier to publish before someone has identified whose expression, identity or labor the output uses.

The law lags partly because it works through narrower categories and slower institutions. Copyright asks who contributed original expression. Publicity and privacy rules ask whose identity or dignity is implicated. A collective agreement allocates consent, compensation and working conditions. Legislatures add targeted protections; courts decide particular disputes on particular records. A model update does not wait for any of these processes. Existing law still applies even where there is no AI-specific statute.

Our hypothetical film, The Last Broadcast, contains a human-written script, generated city exteriors, human editing and a synthetic narrator that resembles a known performer. It will also have a realistic online advertisement and an EU release. Its producer must answer three recurring questions: what can we protect, whose permission or payment do we need, and how must we explain what audiences are seeing? The five objectives above are the practical skills needed to build that clearance file.

U.S. copyright protects original expression fixed in a tangible medium under 17 U.S.C. §102. It does not protect a bare idea, production method or general style. Human authorship is essential: a work generated entirely by a machine without human authorship is not eligible for U.S. copyright. This is a rule about protection for that expression, not a judgment that the work lacks artistic or commercial value. S1, S4

In Thaler v. Perlmutter, the application identified an AI system as the author of A Recent Entrance to Paradise. The D.C. Circuit affirmed the human-authorship requirement on 18 March 2025. The Supreme Court denied review on 2 March 2026. Denying certiorari leaves the lower-court judgment in place; it is not a new Supreme Court merits opinion defining every possible human–AI collaboration. S2

For our film, a generated city shot may have no protectable human authorship in its pixels, while the original script and a sufficiently creative final sequence may qualify. Do not infer that the entire film loses protection because it contains generated material, or that the film's registration necessarily gives exclusive rights in every component.

Suppose our editor generates 40 city clips, selects eight and builds a 90-second sequence around a human-written voiceover. She chooses the order, rhythm, transitions and relationship between the shots and narration, and manually creates an original composite in one frame. The relevant issue is the resulting original human expression, not the number of generations or hours worked.

An independently written narration can be protected as writing. Original selection and arrangement may be protected as a compilation; original compositing or editing may add protectable expression. Under §103, protection for a compilation or derivative work does not extend ownership to unprotected or pre-existing underlying material. A competitor copying the creative sequence presents a different claim from one reusing an unprotected generated clip. Routine trimming, format conversion or purely mechanical correction may add too little original expression to qualify. S4

A concrete administrative example is Zarya of the Dawn. In its 21 February 2023 letter, the Copyright Office recognized the comic's human text and selection/coordination/arrangement, while excluding the Midjourney-generated images themselves. The lesson for our editor is to describe the human contribution accurately and identify AI-generated material in a registration application as required, rather than claim every pixel. Preserve scripts, project files, edit histories and before/after composites as evidence; retaining records does not itself establish eligibility. S3

The brief supplied approximate figures of 250 AI cases, 180 active, 145 copyright and 55 naming OpenAI. A count without a tracker, date and inclusion rules is not a stable fact. Cases may be consolidated, dismissed, settled, appealed or counted separately against multiple defendants. Intellectual property is broader than copyright; a pending motion and an entire pending lawsuit are different units.

As one reproducible snapshot, the World of AI tracker opened on 13 November 2026 listed 203 cases, 111 classified as intellectual property, 42 not yet classified, and 63 naming OpenAI. It named OpenAI most frequently within its dataset. Those are the publisher's dataset figures, not an official global census, and its 203 total is not a count of active cases. The supplied 250/180/145/55 figures could not be reconciled to a named snapshot and should be labeled unverified rather than republished as current statistics. S5

Our producer should read a case update through four questions: who is suing whom; what act is challenged; what stage has the court reached; and what question remains undecided? A complaint states allegations. Surviving dismissal generally means a claim may proceed, not that infringement or causation has been proved. A settlement does not necessarily establish a legal rule. A ruling may bind only a particular jurisdiction or resolve only one issue.

Training-data disputes concern acts such as acquiring books, copying photographs, retaining datasets and using works to train a model. Output disputes concern generated material that allegedly reproduces protected expression. A producer's own release can raise a separate question from the developer's training conduct. A model's availability does not settle either issue.

Fair use is a statutory defense under 17 U.S.C. §107. Courts weigh the purpose and character of the use; the nature of the copyrighted work; the amount and substantiality taken; and the effect on the actual or potential market. A transformative purpose can matter to the first factor, but merely calling a tool innovative or adding new meaning does not finish the inquiry. There is no universal percentage of permitted copying, automatic educational exception, or categorical rule that every training use is lawful. S4

Consider two hypothetical uses in The Last Broadcast. A journalist includes a brief, necessary clip from an AI advertisement while analyzing its deception, with commentary directly engaging the clip. Another producer uses the entire advertisement as attractive filler. Both involve copying, but their purposes, amounts and market effects differ. Neither is resolved by an on-screen credit alone. Market substitution asks whether the use displaces demand for the original or a legally cognizable licensing market, rather than simply whether a new product competes in some broad sense.

A shadow library is a collection that makes works available without the relevant rightsholders' authorization. How a developer obtained a work may be legally distinct from how it later used it. The Copyright Office’s fair-use index records a mixed result in Bartz v. Anthropic (23 June 2025), and a fair-use finding in Kadrey v. Meta (25 June 2025). These are dated district-court outcomes, not a nationwide rule that all AI training is lawful. In Bartz, the ruling distinguished training from acquisition and retention of pirated library copies. In Kadrey, the ruling turned on the evidence presented by the particular plaintiffs, including market harm. Always separate the specific ruling from later settlement or appellate developments. S37 For the film's clearance file, record licensed source assets and vendor assurances, investigate suspiciously recognizable outputs, and do not treat a fair-use argument by a model developer as a release-wide warranty.

Chatbot-harm litigation expands the landscape to alleged wrongful death, self-harm, design defects, failure to warn, privacy violations and nonconsensual sexual deepfakes. Claims involving Character.AI and xAI/Grok must be separated by product, alleged conduct, jurisdiction and procedural stage. A suit alleging that a chatbot contributed to a person's death differs from a suit alleging that an image tool generated sexualized depictions of minors. Both can involve causation and platform defenses; neither allegation is a finding of liability. In Garcia v. Character Technologies, the family of a 14-year-old who died by suicide challenged the chatbot’s design and warnings. The court’s May 2025 order allowed substantial claims to proceed at the pleading stage, taking properly pleaded allegations as true for that motion. It did not establish that the product caused the death. A January 2026 Doe v. xAI complaint challenged Grok’s alleged creation and dissemination of nonconsensual sexualized images; a separate March 2026 complaint by three plaintiffs alleged exploitation involving images of them as minors. These filings show the diversity of harm theories, rather than adjudicated proof of the allegations. S34, S35, S36

Florida's attorney general sued OpenAI and Sam Altman on 1 June 2026, alleging deceptive practices and harms including risks to children. On 28 September 2026, the state sought a temporary injunction concerning model safeguards and minors' access. S38 A request for an injunction is a request for a court order; it does not itself mean the court granted a ban. Use the official complaint to identify the state's claims, and a verified subsequent order to describe what the court actually required. S6

For our journalist covering the film's synthetic narrator, the practical skill is precise attribution: write “the complaint alleges,” “the court allowed this claim to proceed,” or “the court held,” as appropriate. For a producer, a copyright license does not answer product safety, exploitation or deceptive-marketing concerns.

A digital replica or synthetic double is a realistic digital representation identifiable as a particular person's voice or visual likeness, capable of appearing or performing in circumstances that person did not actually perform. A synthetic performer, as used in the relevant union provisions, is a generated asset giving the impression of a natural performer without being recognizable as a particular natural performer. Definitions vary between statutes and agreements; use the applicable instrument's definition. S7

This difference matters in our film. A cloned voice recognizable as a working actor implicates that person's consent and compensation. A fictional generated spokesperson may instead trigger synthetic-performer bargaining and advertising disclosure. Removing the person's name from a prompt does not necessarily remove recognizability. An ordinary edit, conventional effect or authorized adjustment is not automatically equivalent to a new digital performance.

SAG-AFTRA's 2026 TV/Theatrical agreement covers specified productions and signatory obligations; it is not a statute governing every creator worldwide. Its stated term runs from 1 July 2026 to 30 June 2030. Determine the agreement governing the film, the replica category and the production/use date before applying any rule. Commercials, interactive media and other contracts can have different terms. S7

For covered digital replicas, meaningful written consent must identify intended uses with the specificity the agreement requires. The agreement requires clear, conspicuous, use-specific consent in a separately signed or initialed contractual statement or signed writing, as applicable. Consent to creating a replica is not a blanket consent to every later deployment. Employment-based and independently created replicas have different creation and payment rules. A prior release allowing undefined future uses is not a substitute for checking the applicable protections. S9

Illustrative budget arithmetic, not a union rate quotation: suppose a performer's negotiated covered rate is $2,000 per day and a permitted replica use is assessed under the agreement as substituting for two days of performance. The starting performance-payment calculation would be 2 × $2,000 = $4,000 under those assumptions. Creation services, minimum calls, benefits, residuals, use category and bargaining can change the actual obligation. “We paid for the scan” does not establish that we paid for the later performance. State every assumption before quoting a production budget.

The 2026 agreement sets a day-performer minimum for use of independently created replicas and requires residuals when the replica performance would qualify a natural performer for residuals. It also adds protection for replicas created from production photography without a separate scan. Foreign-language dubbing has a distinct new-consent trigger for pictures commencing principal photography on or after 1 July 2027. Verify those dates rather than assume every change began in July 2026. S8

The producer's clearance file should identify whose replica is involved, how it was made, the approved scenes and uses, any altered or later use needing renewed consent, the negotiated payment basis and applicable minimums, residuals and benefit obligations, storage/access controls, and transfer restrictions. A vendor's asset delivery does not discharge the producer's covered obligations.

For employment-based replicas, use compensation depends on the applicable engagement and the work replaced; it is not a universal flat “usual rate” paid for every alteration. The agreement includes negotiated-compensation rules, specific exceptions and separate bargaining for other-picture or other-medium uses. Budget creation and use separately, then check the actual clause rather than treating the illustrative $4,000 as a quoted minimum. S9

The 2026 summary distinguishes two safeguards. An articulable business reason applies when seeking consent to scan a performer for an employment-based replica. The separate synthetic-performer provisions favor human performances, require notice/bargaining and state the commitment concerning significant additional value in human roles. Avoid collapsing them into a single statutory ban on replacing actors. S8

For the film, a request to scan its actor needs its own justified production purpose. A proposal for a wholly generated human-looking lead needs a separate contractual review. Cost avoidance should not be assumed sufficient to demonstrate the required additional value. If synthetic provisions are violated, arbitration damages need not be limited to the compensation a natural performer would have received; that is not an automatic preset award. S9

The agreement also prohibits using a minor performer's replica to depict nudity or simulated sexual activity. The operative text expressly includes aging up or de-aging a minor’s performance for those prohibited purposes. Consent from an adult or guardian is not a general license to bypass child-protection rules. Teach this issue without depicting the prohibited content.

Tennessee's ELVIS Act, effective 1 July 2024, extends state personal-rights protections to voice, including simulations. Its scope can reach specified unauthorized uses/distribution and technology directed at producing a particular person's voice or likeness, subject to statutory conditions and exceptions. It is not strict liability for every general-purpose tool. S10

California AB 2602 targets certain personal/professional-services contract clauses authorizing digital replicas as substitutes for work. A clause can be unenforceable when the statutory conditions are met, including insufficiently specific intended uses and lack of qualifying lawyer or union representation. Labor Code §927 applies to the qualifying new replica performance fixed on or after 1 January 2025. This is a focused contract protection, not a declaration that every replica contract is void. S39 S11

California SB 1050 was signed on 16 September 2026 and is scheduled to take effect on 1 January 2027. It concerns clear, conspicuous disclosure of synthetic performers in covered advertising, including audio. Its chaptered text calls for disclosure near the synthetic performer or in the audio, as applicable, and distinguishes the ad creator’s duty from an advertising medium’s obligation after service of a relevant court order. S33 New York's law, signed 11 December 2025 and effective in June 2026, similarly addresses synthetic performers in ads, with its own exceptions, including audio-only advertising and specified expressive-work promotions. Do not treat either law as a blanket label requirement for every AI-assisted movie. S12, S13

At the federal level, the NO FAKES Act, S.1367, introduced on 9 April 2025, proposes voice and visual-likeness rights. Treat it as pending legislation, not an existing nationwide license requirement. In Lehrman v. Lovo on 10 July 2025, a New York federal court dismissed most federal copyright/trademark theories concerning voice clones while allowing important state-law claims to proceed. A person's vocal identity is different from copyright in a recording of it; the ruling illustrates gaps, not a finding that performers have no legal protection at all. S14, S15, S42

Europe does not supply one general EU-wide voice-and-likeness right lasting 50 years after death. National personality, image, privacy and neighboring-rights rules differ. The Digital Services Act can structure notices and platform obligations concerning illegal content, but does not itself create that asserted duration. S41 The judgment in the Berlin voice-clone case—LG Berlin II, 20 August 2025, 2 O 202/24—illustrates national personality-rights remedies. It awarded €4,000 for two videos, using a hypothetical license-fee valuation of €2,000 per video, based on the case’s facts. That is not an EU tariff. The judgment text was checked in a legal-database reproduction; the official Berlin permalink was identified but could not be retrieved. S32

FTC guidance in May 2026 identified potential civil penalties of $53,088 per violation for platform noncompliance. This is an enforcement figure, not a fixed payout to every victim; civil penalties may be adjusted, and how violations are counted is a legal question. Preserve the date attached to the number. S18

James Strahler II pleaded guilty in April 2026 to offenses including cyberstalking, producing obscene visual representations of child sexual abuse, and publication of digital forgeries. DOJ identified him as the first person convicted under the new Act. On 8 September 2026, he received a total 180-month sentence—15 years—for the combined cybercrime conduct. Do not describe 15 years as the maximum punishment for a single TAKE IT DOWN adult-publication offense. S19

For our film's marketing team, intimate synthetic material is not merely another rights-clearance line. A performer release, platform label and film copyright do not answer whether a depiction or distribution violates intimate-imagery or child-protection law. Escalate a problematic proposed asset before publication; do not circulate harmful examples as training illustrations.

Google's SynthID embeds an imperceptible watermark into generated content, using techniques appropriate to images, audio, video or text. Detection looks for that signal. It does not determine whether every statement is true or whether the depicted performer consented. Detection and robustness have limits; a missing signal does not establish human origin. S26

C2PA Content Credentials use cryptographically signed provenance statements associated with an asset. They can describe origin and editing history and enable verification of signatures and whether protected parts of the asset changed. They are not merely a visible “AI” badge and are not guaranteed to survive every export or screenshot. Provenance reports a documented chain of assertions; a valid signature does not establish the truth of the depicted event. Absence of credentials is not proof of fakery. S27

Both technologies can support a transparency workflow, but neither product name guarantees legal compliance. A provider must meet the actual technical/legal requirements, and a deployer may separately need audience-facing disclosure. Our film should preserve provenance through editing and delivery, inspect the exported master and the platform-transcoded version, and retain the underlying records. A visible label, invisible signal and signed history perform different jobs.

The U.S. does not have a directly equivalent, general federal AI-content marking regime to Article 50 at this research cut-off. That does not mean there are no federal obligations involving AI media: TAKE IT DOWN and existing consumer-protection law are examples. The AI Labeling Act of 2026, S.4915, introduced 24 June 2026, is a proposal; do not describe its planned marking system as already binding. State advertising laws can apply independently. S28, S43

Distribution gates also impose requirements. Sundance's 2027 submissions FAQ requires disclosure of AI use while stating that it does not generally prohibit such technology. This is evidence of a specific festival's policy, not proof that all festivals follow the same rule. Check the actual edition's submission form and rules. S31

YouTube requires disclosure for realistic, meaningfully generated or altered content, with exceptions for nonrealistic or minor changes; its 2026 update makes labels more prominent. Meta's policy uses industry signals and uploader disclosure to apply AI-information labels, while other content rules remain applicable. A platform may label content automatically, but the producer should not assume automatic labeling satisfies every uploader or statutory obligation. Policies change independently of legislation. S29, S30

Fair use: the context-specific limitation/defense evaluated under the four statutory factors in 17 U.S.C. §107; it is not advance permission from a rightsholder.

Transformative use: a use serving a sufficiently different purpose or character relevant to fair-use analysis; new technology, a new aesthetic or a commercial label alone does not settle the analysis.

Shadow libraries: collections distributing or providing access to works without relevant rightsholder authorization; acquisition and subsequent uses can raise distinct legal questions.

Market substitution: displacement of demand for an original work or a legally relevant licensing market, examined on the facts rather than assumed from any competition.

Digital replica / synthetic double: a digital representation recognizable as a particular person's voice or visual likeness; check the governing definition. A synthetic performer can instead be a fictional human-like performer without a recognizable real-person identity.

Content Credentials / provenance: signed information about an asset's origin and history; useful evidence of recorded assertions and changes, not a guarantee of factual truth, permission or human authorship.

Recap: protect original human expression; separate platform contracts from third-party clearance; report litigation by source, date and procedural posture; check recognizable performers and covered labor obligations; apply harmful-media rules to the actual conduct; and meet the destination's marking, disclosure and distribution requirements. Keep the clearance evidence with the master. Educational information, not legal advice.