Do You Need 2257 Records for AI-Generated Adult Content?
Your acquirer's application asks for your 2257 custodian of records. Your platform has never had a performer. Every image it has produced came out of a diffusion model, and the people in them do not exist.
So do you keep records on nobody, post an exemption and hope, or build a records system for a law that may not apply to you?
Operators frustratingly search this exact question every day and find two kinds of answers:
Boilerplate exemption statements copied between AI sites, and, well, silence.
Here is what the statute actually says, where AI-generated content sits inside it, and what you should do regardless of how the legal question eventually resolves.
To be clear, this is general information, not legal advice; the right answer for your platform should come from counsel who knows your pipeline.
Key takeaways
• 18 U.S.C. 2257 requires producers of actual sexually explicit conduct to keep age and identity records for every performer, and a performer is defined as a person. Fully synthetic content with no real human has no performer to record.
• The statute does cover digital images and digitally or computer-manipulated images of an actual human being. The moment a real person's likeness enters the pipeline, you should assume 2257 applies.
• No court or Department of Justice guidance has squarely addressed purely AI-generated adult content. That's a gap, and acquirers fill it with their own requirements.
• Card networks require identity and consent for anyone depicted and apply their rules to generated content, so the practical standard is stricter than the statute whatever the statute means.
What 2257 actually requires
Section 2257 applies to anyone who produces matter containing visual depictions of actual sexually explicit conduct made after November 1, 1990.
For every performer, the producer must inspect a government-issued ID, record the performer's legal name, date of birth, and any aliases or stage names, keep those records available for inspection, and place a statement on every copy, including every page of a website, saying where the records are kept and naming the custodian.
Failing to keep records, falsifying them, or omitting the statement is a felony carrying up to five years for a first offense.
Two definitions matter for AI.
The statute's trigger is actual, not simulated, sexually explicit conduct; simulated conduct falls under the companion section 2257A, which has its own narrower scheme. And a performer is defined as any person portrayed engaging in or assisting in that conduct. Both definitions assume a human being was in front of a camera.
Where AI-generated content falls
The statute's list of covered matter includes a digital image or a digitally or computer-manipulated image of an actual human being. That phrase can be read a couple of ways.
• Fully synthetic output. A model trained on licensed or general data generates a person who does not exist. There is no actual human being depicted, no performer, and nothing to inspect an ID for. On the text of the statute, the record-keeping duty does not attach, which is the reasoning behind the exemption statements many AI platforms post.
• Any real-person input. A user uploads a reference photo, a face is swapped onto generated output, a LoRA or fine-tune is trained on a specific individual, or a creator uses the tool on their own likeness. The result is a computer-manipulated image of an actual human being, and that person is portrayed in the conduct. Assume full 2257 records are required, along with that person's consent.
• Hybrid content. Real performer footage enhanced, extended, or altered by AI remains filmed content of a real performer. Nothing about the AI step removes the original obligation.
Earlier versions of the statute referred to computer-generated images directly; a 2006 amendment replaced that language with the current actual-human-being wording. No published court decision has applied 2257 to purely synthetic adult content, and the Department of Justice has not issued guidance. Operators are working in a true gap.
Why the enforcement picture is also unsettled
The Free Speech Coalition spent more than a decade challenging 2257 and 2257A.
In September 2020 the Third Circuit held that the age verification, record-keeping, and labeling requirements violated the First Amendment as applied to the specific producers who brought the case, and limited relief to those plaintiffs.
The statute remains in force for everyone else, the Department of Justice has not disclaimed enforcement, and the industry has kept complying. For an AI operator, the lesson is that the law is both unresolved and still on the books, which is the worst combination to build a strategy around.
Why the legal answer is not the whole answer
Even if a court eventually confirms that synthetic content needs no 2257 records, your acquirer will still ask for them, or for something that does the same job.
The Mastercard adult content rules require government ID and documented consent for every person depicted in monetized content and apply to content a provider can upload or generate. Visa's Integrity Risk Program expects adult merchants to demonstrate that depicted persons are verified adults who consented. Neither network has published an AI exemption, and acquirers underwriting adult AI and NSFW AI payment processing generally expect a 2257 statement or a well-reasoned exemption statement on the site, a named custodian, and a records process for anything involving a real person.
The bank's concern is simpler than the statute's. It wants to know that no real person appears in your output without verified age and consent, and that you can prove it. Your records system is how you prove it.
What to do now
The practical posture that satisfies counsel and acquirers at the same time looks like this.
• Decide, in writing, whether real-person likenesses are allowed on your platform. If they are not, enforce it technically; our guide on how to prevent deepfakes and NCII covers the controls. A platform that blocks real people has a defensible basis for an exemption.
• Post a 2257 statement that matches that decision. For fully synthetic platforms, an exemption statement should explain that output depicts no actual human being, describe the controls that make that true, and still name a custodian and address for inquiries. Have counsel draft it rather than copying another site's.
• Keep full 2257 records for any real-person pathway. Creator accounts, consented likeness features, and hybrid content get ID inspection, consent forms, and the standard statement on every page where the content appears.
• Maintain provenance records for synthetic content. Model and version, training data statements, generation logs, and classifier results are the AI equivalent of performer records, and acquirers increasingly ask for them under the label of content provenance.
• Name a custodian anyway. A real person, a business address, and a documented retrieval process cost little and answer the application question directly.
• Revisit the position annually or when the law moves. Add it to the same calendar as your VIRP attestation and your acquirer's annual review.
Frequently asked questions
Does 2257A apply to AI-generated content instead?
Section 2257A covers simulated sexually explicit conduct by real performers, with a certification regime for certain producers. Like 2257, it presumes a human performer, so it does not fit purely synthetic content any better. Content involving a real person's likeness should be analyzed under both sections with counsel.
Can I just copy a 2257 exemption statement from another AI site?
You can, and your acquirer will recognize it. A statement that does not describe your actual controls proves nothing and can be worse than none if an auditor finds real-person features it fails to mention.
Will an acquirer approve a synthetic-only platform with an exemption statement?
Often, yes, when the exemption is backed by enforced likeness controls, provenance records, and a named custodian. Approval depends on evidence, not on the statement itself; see our guide to how to prove content moderation for real-time generation.
Next step
2257 was written for cameras and performers, and AI platforms are living in the gap it left. The merchants that keep processing are the ones that treat the gap as a reason to document more, not less. MobiusPay underwrites synthetic and hybrid adult AI platforms, knows which records satisfy acquirers in each case, and offers the same risk management support to traditional adult merchant account clients. If you need a processor that can tell you what the bank will actually ask for, get started and bring your current 2257 statement with you.
Related Articles
How to Prevent Deepfakes and NCII on Your AI Platform
Deepfakes and NCII can end your merchant account overnight. Learn how to block real-person likenesses and meet the TAKE IT DOWN Act 48-hour rule.
How to Write an Acceptable Use Policy for an NSFW AI Platform
Learn how to write an acceptable use policy for an NSFW AI platform, define prohibited content, enforce rules, and prepare for payment partner review.
How to Prove Content Moderation When Your AI Generates in Real Time
Banks want pre-screening, but your model generates on demand. Here is how adult AI platforms prove content moderation and keep processing
