AI Regulation Tracker / US state law
Hawaii's AI likeness law is already in force, with no preparation window
Act 247 makes it unlawful to knowingly publish a realistic AI imitation of an identifiable person without consent in an advertisement. Damages run up to $25,000 per advertisement, and the Attorney General has a separate channel.
What did Hawaii actually enact?
A new chapter of the Hawaii Revised Statutes, built around a single prohibition. Section 2 of the enrolled bill, at the prohibited-acts section, reads:
It shall be unlawful for any person to knowingly publish a realistic digital imitation of an identifiable individual without that individual's consent if the imitation: (1) Is used in connection with an advertisement; (2) Causes harm; or (3) Is used to commit fraud, defamation, harassment, or other criminal acts.H.B. 2137, C.D. 1, Section 2, prohibited acts
Read the structure carefully. Consent is the gate, knowledge is the mental state, and the three conditions are alternatives joined by "or", not a cumulative test. Use in connection with an advertisement is enough on its own. Nobody has to show harm or fraud. In advertising the plaintiff's case reduces to three questions: was the individual identifiable, was the imitation realistic, was there consent.
The Act also carries carve-outs. Parody, satire, commentary, criticism, scholarship, political and educational expression, news reporting, documentary and biographical works, and media merely disseminating third-party content are exempt, subject to the detailed conditions in the statutory text. Commercial advertising is squarely inside.
Why does "effective upon approval" change the calculus?
Most state AI laws arrive with a runway. Signed in spring, live the following January, with two or three quarters for counsel to rewrite talent agreements and vendor terms. Act 247 does not work that way. Section 5 provides that it takes effect upon approval.
So there is no preparation window. Any campaign running a realistic synthetic likeness into Hawaii after July 14 is already under the statute. Work commissioned in June and placed in August is covered, because what the Act reaches is publication, not creation.
One note on the date. Press coverage ran July 15 and 16, and some secondary reporting places the signing on July 13 or 15. The Legislature's measure record and the Governor's Message number both put Act 247 at 07/14/2026. For date questions, rely on the Legislature's measure-status record, which shows Act 247 as approved on July 14, 2026.
How does the $25,000 figure actually work?
The remedy provision offers a plaintiff a choice: actual damages, or statutory damages of up to $25,000 per advertisement. It is a ceiling, not a floor. Injunctive relief is available, punitive damages on proof of malice, and attorneys' fees are recoverable, which changes the economics of a small claim.
The counting unit in the text is the advertisement. The statute provides for up to $25,000 per advertisement as an alternative to actual damages.
Who can bring an action, and on what trigger?
Two channels, with different triggers, and they should not be collapsed. The first is a private right of action. The identifiable individual sues, and the claim turns on consent. An estate may bring it for up to ten years after death, which puts deceased-talent recreations inside a defined window rather than outside the statute.
The second is an action by the Hawaii Attorney General for equitable relief where the harm is widespread. That trigger is breadth, not consent. A campaign reaching a large audience can draw a state action even where no individual has filed suit, and the relief there is equitable rather than a damages award. A tightly targeted ad is a private-claim problem. A broad campaign is both at once.
What does Act 247 look like element by element?
The statute is short, and each element does a distinct job. Laid out side by side, the elements show where an advertising team's exposure actually sits.
| Element | What the enrolled text provides |
|---|---|
| Conduct reached | Knowingly publishing a realistic digital imitation of an identifiable individual without that individual's consent |
| Triggers | Use in connection with an advertisement, or causing harm, or use to commit fraud, defamation, harassment or other criminal acts. The triggers are alternatives. |
| Statutory damages | Up to $25,000 per advertisement, as an alternative to actual damages |
| Other remedies | Injunctive relief, punitive damages on proof of malice, and attorneys' fees |
| Who may sue | The identifiable individual, and the estate for up to ten years after death |
| State channel | The Hawaii Attorney General, for equitable relief where the harm is widespread |
| Exemptions | Parody, satire, commentary, criticism, scholarship, political and educational expression, news reporting, documentary and biographical works, and media merely disseminating third-party content, subject to the conditions in the statutory text |
| Effective date | July 14, 2026, upon approval, under Section 5 |
The advertising trigger is the one that matters most in practice, because it asks nothing about harm. Identifiability, realism and consent carry the whole claim.
What should a US marketing or brand team do about it?
The duty this creates is a documentation duty. If you generate a realistic likeness of a real person for an advertisement running in Hawaii, you need consent on file, in writing, before publication. That is the whole posture in one sentence, and there is no phase-in during which a good-faith gap is tolerated.
Three things follow. Talent releases drafted before generative tools entered the workflow often authorise footage and photographs without addressing synthetic recreation, so the release you hold may not cover what your vendor is producing. Vendor contracts for AI-generated creative should say who warrants consent and who indemnifies when it is missing. Estates are in scope for ten years, so clearance for a deceased figure is a live question.
Geographic targeting is a weaker control than it looks. Digital placement is rarely clean at the state line, and the statute reaches publication. If a national buy touches Hawaii, treat it as in scope. Whether a given imitation is realistic enough, and the individual identifiable enough, stays a lawyer's call.
What Act 247 does not do
It does not require labelling of synthetic content generally, which is the New York approach. It does not regulate model training or the tools themselves, only publication of the output. And it is not a licence regime, so nothing in it sets a rate or a clearance process.
Hawaii enacted an earlier election-deepfake law, Act 191 of 2024. Act 247 is a separate instrument from a different session with a different subject: commercial imitation and consent rather than deceptive political media in an election window, and it carries express carve-outs for political expression, news reporting and satire.
Frequently asked questions
When did Hawaii Act 247 take effect?
On July 14, 2026, the day it was approved. Section 5 of H.B. 2137, C.D. 1 states that the Act takes effect upon its approval, and the Legislature's measure-status record shows it became Act 247 on 07/14/2026 under Governor's Message No. 1349. There is no delayed effective date and no preparation window.
What does the $25,000 figure in Act 247 actually cover?
It is a statutory damages ceiling of up to $25,000 per advertisement, offered as an alternative to actual damages. The counting unit in the statutory text is the advertisement. Punitive damages are available on proof of malice, and attorneys' fees are recoverable.
Who can sue under Hawaii Act 247?
Two separate channels. The identifiable individual has a private right of action, and an estate may sue for up to ten years after the person's death. Separately, the Hawaii Attorney General may bring an action for equitable relief where the harm is widespread. The triggers differ: the private claim turns on the absence of consent, while the Attorney General channel turns on the breadth of the harm.
Last verified: July 28, 2026