Likeness, voice rights and deepfakes

0:00 Likeness, voice and deepfakes

Imagine opening Instagram and watching yourself recommend a slimming tea you've never tried, in a language you don't speak, to your own followers. The voice is yours. The face is yours. You never said a word of it. For creators, this is no longer science fiction. In this lecture you'll learn what makes synthetic media a harmful deepfake, the legal strands that protect people, the clear no-go zones, how to get consent that actually holds up, and how to protect your own identity.

0:36 Why it matters

Why does this matter so much? Because a person's face, voice, name and persona are part of who they are, and for creators, they're also how they earn a living. AI can now copy all of them from a few minutes of footage. Using someone's identity without permission can be unlawful, unethical and deeply harmful. And for businesses, it's a fast way to lose a partner, a customer base and a reputation. The flip side is that consented, transparent synthetic media can be genuinely useful: dubbing, accessibility, scale.

1:14 The deepfake test

So what exactly is a deepfake? Think of it like a forged signature. Signing someone's name isn't the problem when they asked you to sign on their behalf and everyone knows it. The harm comes when it's realistic, and they didn't agree, or people are deceived. Same with synthetic media. Your own consented AI avatar, clearly labeled, is synthetic but not deceptive. A realistic video of a real person saying things they never said, without consent or disclosure, is the harmful kind. Realism plus lack of consent, or deception. That's the test.

1:54 Legal strands

Several legal strands protect people, and they vary by country. Personality and publicity rights protect against commercial use of someone's name, image, likeness and, in some places, voice; Tennessee's ELVIS Act is one example that explicitly covers voice. False endorsement and advertising law, enforced by bodies like the FTC in the US and the ASA in the UK. Defamation. Data protection, because face and voice are personal data, and biometric processing carries stricter rules, including in the UAE and Saudi Arabia. Criminal law on fraud, impersonation and intimate images. And AI-specific rules, like the EU AI Act's deepfake disclosure duty from August twenty twenty-six, and Saudi Arabia's SDAIA deepfakes guidelines.

2:42 No-go zones

Now, the no-go zones. Never create or share synthetic intimate or sexual imagery of a real person without explicit consent. In the US, the TAKE IT DOWN Act made publishing non-consensual intimate images, including AI-generated ones, a federal crime, and many countries have similar offenses. Never create fake endorsements from celebrities, creators, experts or customers. Never make political content showing real candidates saying things they didn't. Never impersonate someone to deceive, like scam calls. And be careful with parody that people could mistake for real, especially about politics, religion or public safety.

3:22 Legitimate use needs

When is using a real person's likeness legitimate? With informed, written consent that covers the specific uses, channels, languages, territory and duration. Script approval rights. Payment. Prohibited uses, like political or adult content. How to withdraw consent. What happens to the voice and face models afterwards, including deletion. For employees, it must be a genuine choice, not a condition of the job, with a clear answer for what happens when they leave. And where content is realistic, you still disclose to audiences. Consent and disclosure are two separate things. You need both.

4:02 Example 1: Lahore tutoring startup

First example. A Lahore tutoring startup wants its founder's voice to narrate lessons in Urdu, English and Punjabi. The founder is happy, but they still do it properly. She signs a consent record covering the three languages, education content only, two years, with her approval on every script and deletion if she leaves. Each lesson says the narration is AI-generated from her voice. When a parent asks whether the founder really recorded everything, the answer is on screen. No awkwardness.

4:37 Example 2 (illustrative): Saudi fashion brand

Second example, a realistic business scenario with illustrative details. A Saudi fashion brand wants to bring back a creator it worked with last year by building an AI avatar from old campaign footage, without contacting her. The agency says no, and explains why. The original contract never mentioned AI replicas. Using her face now implies a current endorsement. And it risks her image rights, data protection law, advertising rules, and SDAIA's deepfake guidance on consent. Instead, they approach her for a new deal: a twelve-month avatar license, script approval, a clear AI label, and a fee. Illustratively, she agreed, promoted the campaign on her own channels too, and the brand got more reach than the avatar alone would ever have delivered.

5:30 Watch me do it: voice clone request

Watch me do it. A client asks: can we clone our sales director's voice for WhatsApp voice notes to customers? I run the five-question check. One, is she real and identifiable? Yes. Two, do we have written consent for this use, channel and term? Not yet. So I stop and draft a consent record: WhatsApp customer updates only, English and Arabic, one year, she approves every script, deletion if she leaves, and it's optional, not part of her job. Three, could customers think she personally recorded each note? Yes. So every note starts with: this is an AI-generated voice message from our team. Four, is it political, intimate, financial or health-related? No. Five, is the record saved? Now it is.

6:22 Common mistakes

Common mistakes. Just a joke deepfakes of friends, colleagues or public figures that escape the group chat. Assuming old contracts cover AI uses they never mentioned. Treating an AI-generated label as a license to depict real people however you like. A label tells people it's synthetic. It doesn't give you permission. Getting a casual verbal OK from an employee. And forgetting deletion, so voice models sit in a vendor account for years after someone leaves.

6:55 Protect your own identity

Finally, protect yourself. Your raw voice and face recordings are training data for anyone who gets them, so keep them in a restricted folder and turn on two-factor authentication everywhere. Read brand contracts carefully. If you see perpetual, worldwide rights to your likeness in all media now known or later developed, push back. Limit term, territory and AI uses, and add approval and deletion rights. Tell your audience how to spot genuine content from you. And report impersonation quickly using platform tools.

7:31 Pushing back: a script

What does pushing back sound like in practice? Here's a simple script. Thanks for the contract. I'm happy with the usage for this campaign. The likeness clause is broader than the project needs, so I'd like to limit it to the named campaign, the agreed channels and twelve months. Any AI replica of my face or voice needs a separate agreement with script approval, a fee and a deletion date. Most brands accept this, because their legal team drafted a wide clause by default, not because they plan to clone you. And if a brand refuses to limit AI rights at all, that tells you something important about the partnership.

8:19 Recap + try this now

Recap. A harmful deepfake is realistic and lacks consent, or deceives. Publicity rights, advertising law, defamation, data protection, criminal law and AI-specific rules can all apply. Some uses are simply off limits. Legitimate uses need specific, written, revocable consent plus disclosure. Try this now: open one brand or creator contract template you use and add an AI likeness and voice section using the consent record in the lesson: scope, term, approval, payment and deletion. Next module, we move to labeling AI content on platforms.