# Rights, authenticity, and deepfakes in the AI content era
In 2023, the film *Rogue One* used a digitally recreated version of Peter Cushing, an actor who died in 1994, to reprise his role as Grand Moff Tarkin. The studio negotiated with his estate. But the technology that made it possible has since become cheap, fast, and widely available. Today any studio, YouTuber, or scammer can generate a convincing synthetic performance of a living or dead person in hours, often without asking anyone.
That shift breaks the old machinery of Hollywood: contracts, residuals, and consent. This lesson walks through the new economics studios are being forced to build.
Three things happened at once.
The tech got good. Generative video models can now produce photorealistic faces and cloned voices from a few minutes of reference material. A "deepfake" (synthetic media that swaps or fabricates a person's likeness or voice) no longer requires a VFX team.
The scale got large. You can generate thousands of variations cheaply. That means a single licensed performance can be reused, altered, and monetized far beyond the original shoot.
The law started reacting. The 2023 SAG-AFTRA strike centered heavily on AI. The resulting contract requires informed consent and compensation before a studio creates or uses a "digital replica" of a performer. This is the single most important framework to understand.
Think of a performer's likeness as a licensable asset, similar to music rights.
Before AI, an actor sold their time on set. Now they can license a *replica* that keeps working after they go home, or after they die. That creates new revenue lines and new fights over who captures the value.
Under the SAG-AFTRA framework, two categories matter:
Employment-based digital replica: created from material captured during a specific job. The studio needs consent for that job, and separate consent to use it elsewhere.
Independently created digital replica: built from other sources (old footage, public appearances). This requires its own explicit consent and bargaining.
The practical takeaway for studios: a signature on a standard contract is no longer enough. Consent must be specific, informed, and scoped. "We can use your face in this film" does not mean "we can use your face in a sequel, an ad, or a video game."
A living actor can negotiate. A dead one cannot. Postmortem publicity rights vary wildly by jurisdiction. Some US states protect them for decades; others barely at all. Studios wanting to resurrect a deceased performer must clear rights with the estate, and increasingly, performers are addressing digital replicas directly in their wills and contracts.
Expect "digital afterlife" clauses to become standard in talent deals, the same way music catalog rights are negotiated today.
If anything can be faked, the scarce asset becomes proof of authenticity. Two approaches are emerging.
A watermark embeds a hidden or visible signal marking content as AI-generated. Google's SynthID embeds an imperceptible watermark into AI-generated images, audio, and video that survives some editing. The weakness: watermarks can be stripped or degraded, and they only work if the generator adds them in the first place.
A stronger approach records where content came from and how it was edited, using cryptographic signatures attached to the file. The leading standard is C2PA (the Coalition for Content Provenance and Authenticity), backed by Adobe, the BBC, Microsoft, and others. You can read the open specification at the C2PA site.
The consumer-facing version is often called "Content Credentials," a kind of nutrition label for media showing the capture device, edits, and whether AI was involved.
Here is what a simplified provenance record conceptually contains:
{
"asset_id": "trailer_final_v3.mp4",
"captured_by": "ARRI Alexa 35",
"edits": [
{ "tool": "DaVinci Resolve", "action": "color grade" },
{ "tool": "GenAI Model X", "action": "background extension", "ai_generated": true }
],
"signed_by": "StudioName",
"signature_valid": true
}The point is not the code. It is that authenticity is becoming machine-checkable. A newsroom or platform can verify that footage came from a real camera and was not synthetically altered, or flag exactly which parts were.
The winners in this era will treat likeness and voice as managed rights portfolios, not one-off contracts. Here is the emerging playbook.
Track exactly what each performer agreed to: which projects, which uses, which time windows, which markets. A performer might allow their replica in a film but forbid political ads or explicit content. This needs to be logged and enforceable, not buried in a PDF.
Several structures are appearing:
There is no settled market rate. Anyone quoting a standard price is guessing.
Studios are beginning to attach content credentials at the point of creation, so every asset carries its edit history. This protects the studio too: if a fake trailer or a fabricated scandal clip circulates, verifiable provenance is the defense.
When an unlicensed deepfake of your talent appears, you need a fast path to remove it. In the US, the proposed federal NO FAKES Act would create a nationwide right against unauthorized digital replicas. As of early 2026 it has not become law, so studios still rely on the state-by-state patchwork plus platform policies. Do not treat any of this as legal advice; consult counsel for your jurisdiction.
Knowledge check
1. The lesson describes the Peter Cushing recreation in Rogue One as one where the studio negotiated with his estate, yet frames it as the start of a broader problem. What concept does this example primarily illustrate?
2. Why does the lesson describe US right of publicity law as creating a 'patchwork'?
3. The lesson compares a performer's likeness to a licensable asset like music rights. What is the key economic implication of this framing combined with generative AI's scalability?
4. Select ALL correct answers. According to the lesson, which factors converged to make rights and authenticity a pressing issue in the AI content era?
Select all the correct answers.
5. Select ALL correct answers. Based on the core term definitions in the lesson, which statements are accurate?
Select all the correct answers.
Not all uses carry equal risk. A rough hierarchy:
Highest risk: using a deceased or non-consenting person's likeness commercially, especially in a state with strong publicity rights, and especially in deceptive contexts (fake endorsements, fabricated news).
Medium risk: reusing a licensed replica beyond the agreed scope. This is a contract problem more than a legal-unknown problem.
Lower risk: clearly labeled synthetic content, fully consented, within scope, with provenance attached.
The pattern is clear: consent plus disclosure plus provenance moves you down the risk ladder. Absence of any one moves you up.
There is a business upside, not just liability management. As synthetic content floods every platform, audiences and advertisers will pay a premium for sources they can verify. A news organization or studio with airtight provenance becomes more valuable precisely because trust is scarce. Authenticity is turning into a competitive moatmoatA lasting edge over competitors: a resource, capability or position they cannot easily replicate, letting a firm earn above-average returns over time.View full definition →.
A studio wants to feature a beloved actor who died in 2010 in a new film.
1. Rights check: Which state governed the actor? Does it protect postmortem publicity, and for how long?
2. Estate negotiation: Secure a license from the rights holder, scoped to this project.
3. Consent record: Document permitted uses (this film only, no ads, no sequels without renegotiation).
4. Compensation: Agree on structure (likely revenue sharerevenue shareThe percentage of total industry sales your company captures in a given period. It measures competitive position relative to rivals in a defined market.View full definition → or usage-based, given the ongoing asset).
5. Provenance: Attach content credentials disclosing the digital replica.
6. Disclosure: Decide how to inform audiences, in credits or marketing.
Skip step one and you may have no right to the likeness at all. Skip step five and you cannot prove you did anything correctly.