2026's AI Disclosure Laws Converge: EU Article 50, California SB 243 and SB 1050, and New York's Synthetic-Performer Law, Compared

Three regimes, one compliance surface

For most of the last decade, disclosing that a piece of content or an interaction involved artificial intelligence was a matter of platform policy and brand judgment, not law. That changed inside a narrow window. Between January and August 2026, three separate lawmaking bodies, the European Union, the State of California, and the State of New York, bring AI disclosure obligations into legal effect, each with its own text, its own scope, and its own penalty schedule.

The convergence is one of timing, not of design. These statutes were not coordinated, and they do not share a common definition of what must be disclosed or when. The EU AI Act reaches across an entire regulated market and governs providers and deployers of AI systems generally. California's laws split the problem in two, one statute for conversational companion chatbots and a separate one for synthetic performers in advertising. New York's law is narrower still, aimed only at synthetic performers and keyed to whether the advertiser has actual knowledge one is used. A business that operates in one jurisdiction may face one of these; a business that markets nationally, or to European users, can face all of them at once.

The practical consequence for a US business is that AI transparency requirements are now a jurisdiction-by-jurisdiction matter with no federal floor to standardize them. Reading the three regimes side by side, on their triggers, thresholds, and penalties, is the only way to see where an organization's exposure actually sits.

EU AI Act Article 50: transparency as a market-wide obligation

The European Union's Artificial Intelligence Act, Regulation (EU) 2024/1689, is the broadest of the three regimes. Its Article 50 sets out the transparency obligations that attach to specific categories of AI system regardless of risk classification: systems that interact with people must disclose that the person is dealing with an AI, and content that is artificially generated or manipulated, including so-called deepfakes and AI-created text published to inform the public on matters of public interest, must be marked as such.

These Article 50 obligations become enforceable on August 2, 2026. The enforcement teeth are significant: the AI Act's penalty structure allows fines up to 15 million euros or 3 percent of a firm's total worldwide annual turnover, whichever is higher, for the relevant categories of infringement. Unlike the two US state laws below, Article 50 is not limited to advertising or to a single conversational format. It is a horizontal transparency rule that applies to the systems themselves.

The reach matters for a US business because the AI Act is extraterritorial in effect: where an AI system's output is used inside the European Union, the obligations can apply to a provider or deployer located outside it. A US firm that offers an AI chat interface to European visitors, or that publishes AI-assisted informational content read in the EU, is inside the conversation whether or not it has a European office.

California SB 243: disclosure for companion chatbots

California took a narrower, conversation-specific route with SB 243, which addresses companion chatbots, AI systems designed to sustain ongoing, relationship-like conversations with a user. The statute took effect on January 1, 2026, and requires an operator to disclose to the user that they are interacting with an artificial intelligence rather than a person.

SB 243 also carries a heightened obligation for interactions with known minors: where the operator knows the user is a minor, the disclosure must be repeated at regular intervals, on a cadence documented in the research underlying this piece as every three hours. This is a design distinction worth noting. SB 243 does not treat disclosure as a one-time notice buried at the start of a session; for the users it most wants to protect, it treats disclosure as something that has to recur so that a vulnerable user cannot forget, mid-conversation, that they are talking to a machine.

California SB 1050 and New York S8420A: the synthetic-performer trigger

The second California statute, SB 1050, moves the disclosure obligation into advertising. It addresses synthetic performers, AI-created depictions of a person used in an advertisement, and requires disclosure that the performer is not real, using sample wording the statute itself specifies. SB 1050 does not set its own dollar penalty; a violation is instead defined as a violation of the state's existing false-advertising statute and is enforced under California's Unfair Competition Law (Business and Professions Code Section 17200 et seq.), which caps civil penalties at $2,500 per violation, sought by the Attorney General or a local prosecutor rather than through a private right of action.

New York's S8420A regulates the same object, the synthetic performer in advertising, and is triggered the same way SB 1050 is: by the presence of a synthetic performer in the ad, not by any percentage of synthetic content. What S8420A adds is a knowledge requirement, the disclosure obligation attaches only once the advertiser has actual knowledge that a synthetic performer is included, and a carve-out for expressive works (film, television, streaming, video games) where the synthetic performer's use is consistent with its use in the underlying work. It takes effect on June 9, 2026, and sets its own fixed penalties: $1,000 for a first violation and $5,000 for each subsequent one.

Reading the three side by side: trigger, threshold, penalty

Laid alongside one another, the three regimes diverge on every axis that matters operationally.

On the trigger: Article 50 is triggered by the category of AI system (interactive systems, generated or manipulated content). SB 243 is triggered by the format of the interaction (a companion chatbot). SB 1050 and S8420A are triggered by a use case (a synthetic performer in an advertisement).

On the threshold: none of the three sets a proportion-of-content test. Article 50 and SB 243 attach to the system or interaction itself, and both SB 1050 and S8420A attach to the presence of a synthetic performer in the ad, not to what share of the ad is AI-made. Where they diverge is scienter and scope: S8420A only fires once the advertiser has actual knowledge of the synthetic performer, and it carves out advertising for expressive works consistent with their underlying use; SB 1050 attaches to the act of creating and publishing the ad and prescribes the disclosure's exact wording and placement.

On the penalty: the EU regime is denominated in millions of euros or a percentage of global turnover. S8420A sets its own fixed dollar penalty, 1,000 dollars for a first violation and 5,000 dollars for each subsequent one. SB 1050 sets no penalty of its own; a violation is enforced under California's general Unfair Competition Law, which caps civil penalties at 2,500 dollars per violation and reserves enforcement to the Attorney General or a local prosecutor rather than a private plaintiff. The gap between a percentage-of-turnover fine and a capped per-violation civil penalty is not a rounding difference; it changes which regime dominates a given firm's risk calculus.

The federal backdrop: no single statute, but the FTC is already here

None of this sits on top of a federal US AI-disclosure statute, because there is not one. What exists federally is a body of Federal Trade Commission rule and doctrine that already reaches AI-related deception without naming a disclosure obligation the way the state laws do.

Two FTC instruments are the most relevant backdrop. The revised Endorsement Guides, 16 CFR Part 255, effective July 26, 2023, extended the definition of an endorser to include AI-created or virtual personas and set a significant-minority standard for when a material connection must be disclosed. Separately, the Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, effective October 21, 2024, made fake-review practices, including AI-created fake reviews, a rule violation carrying civil penalties up to 51,744 dollars per violation.

The FTC also opened an enforcement lane specifically against AI-washing under Operation AI Comply, launched in September 2024, targeting deceptive claims about AI capability. That lane, however, should be read with care: the durability of the AI-enforcement doctrine across administrations is not settled, and the FTC's own reversal of a prior AI-related consent order in December 2025 is direct evidence that federal AI-enforcement posture is contingent rather than a fixed floor. The state disclosure statutes, with their explicit text and effective dates, are the harder-edged obligations of the three-regime picture.

What a nationally selling business must actually track

For an owner-operated or small-team business that markets across state lines, the practical operational read is that these regimes stack rather than substitute. Satisfying one does not discharge the others, because each answers a different question. A companion-chatbot disclosure does not cover a synthetic performer in an ad; a synthetic-performer disclosure that satisfies New York's actual-knowledge trigger says nothing about a European visitor interacting with an AI chat widget on the same site.

There is a further layer for any firm whose delivery structure crosses borders. India's Digital Personal Data Protection Act, 2023, moved into an operational, phased compliance rollout after its implementing Rules were notified in November 2025, with the phase-in concluding in May 2027 and a Data Protection Board empowered to levy penalties up to 250 crore rupees for significant breaches. Raveneye Global is an Indian company, RavenGroup Global Tech Private Limited, that delivers to US clients, so this is a live part of the same compliance surface rather than a distant one.

The point of the comparison is not to prescribe a specific compliance program, which is a matter for qualified counsel in each jurisdiction. It is to make visible that AI disclosure has become a real, multi-jurisdictional legal object in 2026, and that the surface a nationally selling business has to watch is wider than any single headline about a single law suggests.

Reading the evidence

The statutory text and effective dates cited here are established: they come from the regulations and legislative instruments themselves and from legal-tracking coverage of them. What deserves an explicit caveat is that several of these obligations take legal effect within the same 2026 calendar year in which this is written, which makes them fast-moving. Effective dates can shift, guidance interpreting ambiguous terms is still emerging, and the durability of the federal AI-enforcement doctrine is genuinely uncertain rather than merely cautious to say so.

That is the correct posture for a domain like this one. The direction is clear and the hard statutes are real; the precise contours of enforcement are still settling. A business is better served by knowing which regimes exist and what each one turns on than by a confident prediction about how any one of them will be applied.