What happens if your AI campaign triggers a lawsuit?

When an AI-generated face shows up in a national campaign, three parties have legal exposure: the brand, the agency, and the platform that hosts the placement. The model vendor, the company that made the generator, is almost always shielded by terms of service that disclaim downstream liability. So the question isn't whether AI-image lawsuits will hit. It's where they hit, and what happens next.

Step 1: A complaint arrives

Most cases start when a real person sees an ad. They recognize themselves. They don't have to look exactly like the depicted face, they have to be plausibly identifiable. They contact a lawyer. The lawyer drafts a cease-and-desist that names the brand and the agency, demands withdrawal of the campaign, demands documented disclosure of how the image was generated, and reserves rights to seek damages.

The first 72 hours after a cease-and-desist arrive are the worst. The campaign is still running. Internal counsel needs to assess whether to pull it preemptively or hold the line. Pulling is expensive, the media spend is sunk. Holding is risky, every day of continued use compounds damages.

Step 2: The forensic question

The defense almost always begins with the same question: where did this image come from? In a traditional photoshoot, the answer is the contact sheet, the model release, the studio invoice. In an AI-generated workflow, the answer is often: a vendor prompt, possibly a fine-tune, possibly a reference image, and somewhere upstream, a foundation model trained on data with no per-image consent.

This is where the legal trouble compounds. The brand cannot prove the image is "fictional," because the underlying model was trained on real faces. The plaintiff doesn't need to prove the brand intentionally used their likeness, only that a reasonable observer would identify them. The standard is recognizability, not exactness.

Step 3: Damages calibration

In Germany, the dominant framework is "lizenzanaloge Schadensberechnung", damages calibrated to what a license would have cost, plus a multiplier (typically 1.5x to 3x) for unauthorized use. In a national OOH campaign, that license-equivalent fee is not nominal. Add legal costs, the cost of pulling the campaign mid-flight, the brand-trust hit, and you are looking at six- to seven-figure exposure on a single execution.

In the US, right-of-publicity claims layer on top. State statutes vary, but California, New York, and Illinois are particularly aggressive. Illinois's Biometric Information Privacy Act has produced settlements in the hundreds of millions. Federal preemption questions are unresolved.

Step 4: The settlement decision

Most AI-image cases settle. Trial is expensive, public, and the law is moving. Settlement typically combines: campaign withdrawal, a corrective statement (sometimes mandatory, sometimes negotiated), payment of damages, and a non-disclosure agreement. The brand pays. The agency may indemnify. The model vendor watches from a safe distance.

Where licensed identity changes the math

A licensed identity converts an open-ended likeness risk into a defined contract. The depicted person consented to specific commercial use, on specific terms, with a documented audit trail. A complaint becomes a contract question, "did you stay within the scope of the license", instead of a rights question, "did you have rights at all."

That shift has two effects. First, the cease-and-desist often does not arrive at all, because the depicted person is the licensee, they were paid, they have a record, they generally do not file. Second, if a third party complains anyway (a different person who claims resemblance), the brand can produce a contract that demonstrates due diligence and bounded use. That is a far stronger defensive posture than "we used a generator."

Practical implication

For an in-house counsel reviewing AI-generation workflows, the ask is simple: every published asset should map to a license ID. If a complaint arrives, the answer is a file, not an investigation. The cost of integrating a license layer is small. The cost of not integrating it is variable and expensive.