The rules that govern AI in healthcare are moving faster than almost any other part of the field, and they differ sharply by jurisdiction. This page tracks the current status of the major instruments — what is in force, what is only proposed, and the exact dates that matter — with every row tied to a primary regulator source. It is a status tracker for orientation and general information; confirm the live text with the regulator before acting. As of July 2026.
The status table at a glance
| Jurisdiction / instrument | Status | Key date | Source |
|---|---|---|---|
| EU AI Act — Article 4 AI-literacy duty | In force | Applies since 2 Feb 2025 | 1 |
| EU AI Act — high-risk regime (Annex III, Art. 6(2)) | Enacted, awaiting application | Set to apply 2 Aug 2026 | 1 |
| EU Digital Omnibus — Annex III high-risk deferral | Proposed, not yet adopted | Would move to 2 Dec 2027 | 2 |
| US FDA — PCCP marketing-submission guidance | Final | Fed. Register Dec 2024 | 3 |
| US FDA — lifecycle (TPLC) guidance | Draft | Issued Jan 2025 | 4 |
| UK MHRA — AI Airlock sandbox | Pilot + Phase 2 complete | Reports Oct 2025, Jul 2026 | 5 |
| WHO — large multi-modal model guidance | Advisory | Published 18 Jan 2024 | 6 |
| Saudi Arabia — SFDA MDS-G010 | Binding, in force | Guidance MDS-G010 (v. 2023) | 7 |
| Australia — TGA AI consultation | Report published | Jul 2025 | 8 |
| FDA / Health Canada / MHRA — GMLP principles | Published | Oct 2021 | 9 |
Each row is unpacked below with its exact wording and effective date.
European Union: what is live, and what is in flux?
The EU AI Act (Regulation (EU) 2024/1689) applies in stages. The first stage that touches almost every organization using AI is already live: the Article 4 AI-literacy duty, which requires providers and deployers to ensure staff and others operating their systems have a sufficient level of AI literacy, has applied since 2 February 2025 1.
The stage most healthcare builders watch is the high-risk regime. Many clinical decision-support and diagnostic tools fall under Annex III, and under the enacted Regulation the obligations for stand-alone high-risk systems are set to apply from 2 August 2026 1.
Here is the nuance that matters most, and the reason this page carries a ninety-day cadence. On 19 November 2025 the European Commission proposed a Digital Omnibus on AI that would defer stand-alone Annex III high-risk obligations to 2 December 2027 (and to 2 August 2028 for high-risk AI embedded in regulated products). The proposal advanced quickly — the European Parliament approved the agreed text on 16 June 2026 — but as of July 2026 it has not been formally adopted and published in the Official Journal 2. Until it is, the enacted date stands: in law, the 2 August 2026 timeline still applies. Treat the deferral as likely but not yet certain, and watch for the Official Journal notice.
United States: where does FDA guidance stand?
The FDA regulates AI when it meets the device definition, and it has issued two central guidance documents.
The first is final: Marketing Submission Recommendations for a Predetermined Change Control Plan (PCCP) for Artificial Intelligence-Enabled Device Software Functions. Its underlying Federal Register notice dates to December 2024, and the current fda.gov guidance page carries an August 2025 date 3. A PCCP lets a manufacturer pre-specify certain model changes so it can update an AI-enabled device without filing a new marketing submission each time.
The second is still a draft: Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, issued in January 2025 and, as of July 2026, not yet finalized 4. It is the more comprehensive of the two, covering design, validation, transparency, and post-market monitoring across the total product lifecycle (TPLC). Because it remains open, its recommendations can shift before they settle.
United Kingdom: a sandbox, not yet a statute
The UK's approach to AI as a medical device runs partly through the MHRA AI Airlock, a regulatory sandbox where developers and the regulator test how novel or higher-risk AI can operate safely within the existing device framework. The pilot and a second phase are both complete — the Pilot Programme Report was published in October 2025 and the Phase 2 Programme Report in July 2026 — and a third phase is in design 5. The Airlock is a learning mechanism that feeds future rules rather than a binding regime in itself, so read it as a signal of direction.
What role does the WHO play?
The World Health Organization issued Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models on 18 January 2024, with more than 40 recommendations for governments, developers, and providers 6. It carries real influence over national policy, but it is advisory: it sets no market-authorization requirement on its own.
Other jurisdictions in brief
Binding device rules live with national regulators, and several have moved early:
- Saudi Arabia (SFDA): MDS-G010 sets binding requirements for manufacturers of AI/ML medical devices to obtain Medical Device Marketing Authorization — one of the region's earliest binding regimes, per a peer-reviewed analysis in Asian Bioethics Review (2024) of the 2023 guidance 7.
- Australia (TGA): regulates AI under a technology-agnostic device framework; its consultation Clarifying and strengthening the regulation of Artificial Intelligence (AI) ran in late 2024, with an outcomes report published in July 2025 and further work approved 8.
- International harmonization: Good Machine Learning Practice guiding principles were released in October 2021 jointly by the FDA, Health Canada, and the MHRA — an early tri-regulator step still cited across markets 9.
How to read these numbers
Four cautions travel with every row. First, status is perishable: a "draft" becomes "final," and a "proposed" date becomes law or lapses — the EU high-risk deadline is the clearest live example. Second, guidance is not legislation; an FDA guidance or a WHO recommendation shapes expectations but binds differently from a regulation or a statute. Third, scope varies: the same tool can be a regulated device in one market and outside the device definition in another. Fourth, effective dates are staggered within a single law — the EU AI Act alone has separate dates for literacy, general-purpose models, Annex III systems, and product-embedded systems.
We revisit this page every ninety days and whenever a tracked instrument changes status, scope, or date. For adjacent numbers, see our tracker of FDA-cleared AI devices by year and specialty and the AI scribe adoption statistics.
Sources and method. Every row is drawn from a primary regulator source — EUR-Lex for the EU AI Act, fda.gov for FDA guidance, gov.uk for the MHRA, who.int for the WHO, sfda.gov.sa for the SFDA, and the TGA consultation portal — listed and numbered above. Where a date reflects a publishing or notice event rather than a statutory effective date, we say so in the row. Dates and statuses are current as of July 2026.