Brands use AI generated people as fake customers on social media. The synthetic image is the decoy; the non disclosure agreement that silences the contractors who know it is fake is the actual mechanism.
The synthetic influencer is the visible weapon. The non-disclosure agreement is the concealed one.
A Guardian investigation into brands deploying AI-generated people to promote products on social media found that some content creators producing the imagery are reportedly asked to sign NDAs barring them from discussing the work. That single fact reframes the entire story. Every AI-influencer system has exactly one structural vulnerability: a real human who knows it is fake. Brands have identified that point of failure and contracted around it. The synthetic persona cannot speak. The contracted throat is paid not to.
The visible debate, about disclosure, transparency, and labeling, assumes the consumer is the party being deceived. They are. But the consumer is also the party with the least leverage. The parties with leverage are the platforms, the regulators, and the contractors. Of those three, the regulators have already done more than the coverage acknowledges.
Disclosure rules exist. They just are not the AI-specific rules brands ask the press to invoke. The U.S. Federal Trade Commission's Endorsement Guides require clear and conspicuous disclosure of material connections between advertisers and endorsers, and have been read to apply to undisclosed AI-generated endorsements. The European Union's AI Act, through Article 50, obliges providers of generative AI systems to mark outputs in a machine-readable way that detects artificially generated content. The EU's Digital Services Act adds ad-transparency obligations on very large online platforms. State-level unfair and deceptive practices statutes in the U.S. give enforcers and private litigants a second lane. The "regulatory gap" cited in coverage is, on inspection, only partly a gap. It is mostly a case-filing gap.
What is genuinely absent is a rule that addresses the structural coercion at the center of this system: the contract that suppresses the testimony of the person who could otherwise prove the deception. That is not a labeling question. It is a question about whether the operator who produces a synthetic endorsement can be forbidden from contracting away the speech of the human who built it. Speech-suppression clauses in advertising contracts are not new, but their application to AI-generated endorsements introduces a new category of harm. The buyer cannot tell the endorser is not real, and the only witness who could tell them has been paid not to say.
The platform layer is the second-order lever. Meta's advertising standards require disclosure of photorealistic imagery that has been digitally created or altered. Whether the platform's enforcement matches its stated policy is a separate, enforceable question. Distribution is the actual veto surface for the AI-influencer business. A platform that refused to carry undisclosed AI endorsements would collapse the market faster than any labeling rule.
The follow-on effect for real influencers is straightforward. Synthetic personas compete with human creators on price and never sleep. As synthetic endorsement scales, the floor for what a real creator can charge for sponsored work drops, and the only creators with leverage are those whose personal brand is large enough to demand a non-suppression clause. The market does not reward disclosure. It rewards scale. The NDA is what makes the scale viable without the disclosure cost.
What to watch: any FTC enforcement action that explicitly cites a non-disclosure clause as an unfair practice in an AI-endorsement context, any EU AI Act Article 50 enforcement notice against a marketing deployment, any platform that updates its policy to require affirmative labeling of synthetic endorsements, and any contractor who breaks ranks. The structural leverage in this market sits at those four pressure points, not in the synthetic face.