AI transparency · five jurisdictions
The AI transparency regimes, on one timeline
Five laws now tell operators to mark, label, disclose and remove. Not one of them says how to prove you did.
01 / The five regimes
The five regimes
In the order they took effect. Each links to a full reading of the statute — the duties, the thresholds, the penalties, and the evidence question each one raises.
Two labels: one a person can see, one written into the file. Binds generators, dissemination platforms, app stores and publishing users — the only regime here that regulates the whole distribution chain. Actively enforced since November 2025.
Advance notice that a service runs on AI, a label on generative output, and a heightened duty where output is hard to tell from reality. Binding now; the regulator has deferred enforcement, which is a policy rather than a provision.
Labelling and embedded provenance for synthetic media, user declarations on upload, and removal on notice against four separate clocks running from two hours to seven days. No fine attaches — an intermediary that fails simply loses its safe harbour.
A latent disclosure carrying five data elements, a free public disclosure verification tool, and 72 hours to terminate or notify a licensee that has made the system non-compliant. No user threshold since SB 1000. $5,000 per violation, and each day counts separately.
AI interaction notices and machine-readable marking by providers; emotion recognition, biometric categorisation, deepfake and public-interest text disclosures by deployers. Not deferred by the Digital Omnibus, though the marking duty has a transitional to December for systems already on the market.
| Law | Applies from | Status | Penalty | Page |
|---|---|---|---|---|
| China — CAC labelling Measures | 1 Sep 2025 | In force; enforced since Nov 2025 | None in the Measures; administrative sanctions | Read |
| South Korea — AI Framework Act | 22 Jan 2026 | Binding; enforcement deferred by policy | KRW 30m cap | Read |
| India — IT Amendment Rules, 2026 | 20 Feb 2026 | In force | No fine; loss of safe harbour | Read |
| California — SB 942 | 2 Aug 2026 | Operative | $5,000 per violation, per day | Read |
| EU — AI Act Article 50 | 2 Aug 2026 | Applies; not deferred | Art. 50 tier: up to €15M or 3% | Read |
02 / Same day
Two of them start on the same day
On 2 August 2026, California's AI Transparency Act becomes operative and the EU AI Act's Article 50 begins to apply. California moved its own date to match Brussels.
Five prescribed data elements, a public disclosure verification tool, and a 72-hour terminate-or-notify rule for licensees.
No prescribed data elements and no detection-tool duty.
Treating them as one project is the first mistake. One evidence discipline can answer both. One implementation cannot.
03 / Upcoming dates
What comes next
Only dates our regulation pages state.
The transitional closes. Providers of synthetic audio, image, video or text systems on the market before 2 August 2026 must meet the marking duty.
Duties begin for large online platforms and GenAI hosting platforms.
Pending, with no stated date: the end of Korea's guidance period (at least one year from 22 January 2026; a policy announcement), and India's second amendment (consultation closed May 2026; not notified).
04 / In common
What they have in common
Read side by side, these regimes disagree about almost everything — who is bound, which method, what it costs. They agree on one thing none of them writes down.
The model developer, the publisher, the distribution platform, the app store, the end user.
China prescribes a metadata schema down to the field names; the EU prescribes an outcome and names no technology.
From a flat cap worth about twenty thousand US dollars to penalties measured in percentages of worldwide turnover.
Embedding a disclosure, showing a notice, applying a label, verifying a file, terminating a licence, removing content inside a window — each happens once, in your systems, and leaves no durable trace unless you deliberately make one.
Two come close to a record duty: Korea's decree requires the basis for high-impact measures to be retained for five years; China's Measures require six months of logs identifying who received unlabelled output. Neither says whether that evidence may be edited in the interim.
05 / Compare
The same duty, five ways
Who must mark AI-generated content, how, from when, and what happens if they don't — one row per law.
06 / Where Planisphere fits
What Planisphere does not do
The same boundary applies on every page here, in each regime's own vocabulary. Planisphere is an evidence layer. It sits beside your systems, never inside the act.
It does not mark, label, watermark, embed disclosures, or write provenance metadata. Those acts are the duty, and they are yours.
It does not host, distribute, detect or remove content, classify your system, or decide whether an exemption applies.
In order, at the moment you performed them, in a form you cannot quietly revise afterwards.
It does not certify compliance or make a legal determination. No vendor can, in any of these jurisdictions. A record is evidence a third party can check; it is not a finding that you complied.
SOURCES
Go back to the source
Primary text for the China page.
Primary text for the Korea page.
Primary text for the India page.
Primary text for the California page.
Primary text for the Article 50 page.
RELATED
Related: AI content labeling laws, compared · EU AI Act Article 50 · California SB 942 · China AI labelling Measures · India IT Rules on synthetic media · Korea AI Framework Act · Regulation library