AI advertising is moving from a creative experiment to a rights-management problem. On Aug. 25, the company that owns Carl Sagan’s intellectual-property rights sued Luma AI in federal court, alleging that Luma used a clip of Sagan’s voice from Cosmos in social posts promoting its Ray 3.14 generative-video platform. The complaint says the clip appeared on Facebook, X, Instagram and other accounts, and that the use both infringed copyright and falsely implied an endorsement by Sagan’s estate, according to Reuters’ report on the case.

The case, Druyan-Sagan Associates Inc. v. Luma AI Inc., is not a ruling on the merits. It is, however, a sharp warning for every brand, agency and creator using AI-generated video, voice or likenesses in paid media: “the model made it” is not a clearance process.

The Sagan lawsuit turns a familiar creative device into a legal claim

The disputed line is Sagan’s famous sentence, “If you wish to make an apple pie from scratch, you must first invent the universe.” Ann Druyan, Sagan’s widow and owner of the plaintiff, said her late husband’s most cherished principle was his credibility and that she was “aghast at Luma AI’s misappropriation of his legacy,” Reuters reported. The complaint says the estate sent a cease-and-desist letter in February; Luma reportedly characterized the use as “de minimis” and said it would not confuse consumers.

The estate is seeking unspecified damages and an order blocking Luma from using Sagan’s work. Reuters reports that Luma is valued at more than $4 billion and that Ray 3.14 was the product being promoted. For marketers, the important detail is not the size of the defendant. It is the alleged link between an expressive asset, a recognizable voice and a commercial message that could make audiences believe a real person or estate approved the product.

AI tools do not transfer the clearance burden

Platform terms often make the division of responsibility explicit. Luma’s current terms say users are “solely responsible” for inputs and resulting outputs, including their commercialization, and that Luma does not verify ownership or legality of user inputs. The same terms warn that outputs are not guaranteed to be original, unique or free from resemblance to third-party works, and tell users to review and validate generated content before distribution. Read the full language in Luma AI’s Terms and Conditions.

That is the operational shift leaders should understand. A generative tool can accelerate production, but it cannot tell your team whether a prompt contains a protected recording, whether a face or voice was authorized, or whether an edit creates an implied endorsement. A vendor’s indemnity, warranty or commercial-use tier may help, but none of those substitutes for documenting rights before an ad goes live.

Voice and likeness can change the ad’s meaning

Traditional clearance asks whether a team has permission to use a song, image, quote or video. AI advertising adds a second question: does the output make a real person appear to speak, approve or participate? That matters even when the campaign never says “endorsed by.” A familiar voice placed next to a product can create that impression on its own.

The Federal Trade Commission defines an endorsement as an advertising message consumers are likely to believe reflects someone else’s opinions or beliefs. Its guidance says endorsements must be honest and not misleading, and that advertisers and agencies remain responsible for reasonable training, monitoring and review of promotional activity. The FTC’s current guidance is available at ftc.gov/business-guidance/resources/ftcs-endorsement-guides-what-people-are-asking.

The U.S. Copyright Office has separately identified digital replicas as a central AI policy issue and recommended a federal digital-replica law in Part 1 of its AI report. The regulatory direction is still developing, but the business lesson is immediate: treat voice and likeness as controlled brand assets, not as ordinary raw material. See the Copyright Office’s AI initiative.

A practical AI ad clearance checklist

Before publishing an AI-assisted campaign, require four approvals:

  1. Source rights: Record where every uploaded clip, image, script, voice sample and reference asset came from. Keep the license, permission or public-domain rationale with the project.
  2. Identity rights: Ask whether a reasonable viewer could recognize a person, voice or character. If yes, obtain written consent that covers synthetic generation, paid advertising, territories, duration and edits.
  3. Meaning and disclosure: Review the full ad, not just individual frames. If the creative could imply sponsorship or a testimonial, stop and escalate to legal and brand leadership.
  4. Output evidence: Save prompts, source files, model and tool versions, approvals, edits and final exports. That provenance packet is valuable if a platform, claimant or regulator asks how the ad was made.

Marketing leaders should also update agency scopes and creator contracts. Specify who clears inputs, who approves outputs, who monitors live placements and who pays if a claim arrives. Run a small paid test with an original, fully controlled voice or spokesperson before scaling a high-risk concept.

AI can make a campaign faster. It cannot make an unauthorized endorsement safe. Real Internet Sales helps businesses build AI-ready content and digital marketing systems that protect both visibility and trust. Call 803-708-5514 or visit realinternetsales.com.

Sources