One jurisdiction, followed all the way down: the regulation, the
Commission layer, and what Berlin, Paris and The Hague do with it.
Signed webhooks push changes into your systems, a REST API and an MCP
server answer the rest. Every answer carries its primary source.
"applicable": true,
"obligations": [
"Risk management system across the full lifecycle (Art. 9)",
"Automatic event logging (record-keeping) capability (Art. 12)",
…
],
"deadlines": [
"2 December 2027: Chapter III, Sections 1, 2 and 3 … apply to
AI systems classified as high-risk pursuant to Art. 6(2) and Annex III …",
…
],
"sources": ["https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1744"]
The call runs without a key: read access, at most 10 results per answer. A free beta key removes the cap.
Pipelinechecking…
Sources live…
Events on record…
Last ingest…
Dates ahead
The deadlines as they stand after the Digital Omnibus.
2026-12-02
Two new prohibitions in Art. 5 apply. End of the marking transition for generative systems placed on the market before 2 August 2026 (Art. 50(2), Art. 111(4)).
2027-08-02
General-purpose AI models from before 2 August 2025 must comply (Art. 111(3)). National sandboxes operational (Art. 57(1)).
2027-12-02
High-risk rules apply to Annex III systems (Art. 113).
2028-08-02
High-risk rules apply to AI in Annex I products (Art. 113).
Dates re-checked against the Official Journal on 17 November 2026. Set by Regulation (EU) 2026/1744, OJ L, 2026/1744, 24.7.2026. The full timeline with the provision behind each row is in the docs.
The pipeline sweeps 22 official sources, the fastest every 30 minutes. Regulators
publish a handful of items a week, so the dates are their
official publication dates; the sweep is what guarantees nothing newer
exists that we have not seen. Pulled from the public feed as this page
loads, no key involved.
Subscribe to topics like gpai, high-risk,
standards, national-de. Matching updates
arrive as HMAC-signed JSON with exponential-backoff retries. Nothing
to poll, nothing to remember to check.
SHA-256 signature per delivery
Replay-safe event IDs
Receiver examples for 6 platforms
02
REST API
Query the whole record. Filter by source, severity, topic or article
reference. OpenAPI 3 documented, built for backfills, audit trails
and dashboards you own.
Stable event schema, versioned
EN and DE payloads side by side
Cursor pagination
03
MCP server
Drop-in tool for Claude, Cursor or any agent stack. Code scanners
find AI usage in your repo. This tells your agent which obligation
that usage triggers, and cites the regulation for it.
query_ai_act_updates
get_obligation_for_role
summarize_changes_since
Sample webhook payload
What lands at your endpoint.
POST /your-endpoint
{
"id": "evt_2026_1112_001",
"type": "guidance",
"source": "EU_AI_Office",
"topics": ["transparency"],
"severity": "medium",
"title_en": "Commission guidelines on the transparency obligations of Article 50 published",
"title_de": "Leitlinien der Kommission zu den Transparenzpflichten nach Artikel 50 veroeffentlicht",
"summary_en": "The Commission published non-binding guidelines on Art. 50 in the Official Journal (C/2026/6005): marking and detection of synthetic content, deep fake disclosure, exceptions.",
"summary_de": "Die Kommission hat unverbindliche Leitlinien zu Art. 50 im Amtsblatt veroeffentlicht (C/2026/6005): Kennzeichnung und Erkennung synthetischer Inhalte, Offenlegung von Deepfakes, Ausnahmen.",
"url_official": "https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52026XC06005",
"ai_act_articles": ["Art. 50"],
"published_at": "2026-11-12T00:00:00Z",
"language_versions": ["en","de","fr","it","es"],
"disclaimer": "Informational content only. Not legal advice."
}
Not just what changed
What applies to you, today.
A feed only helps on the days something happens. Most of the time the
real question is different: given what we ship and the role we play,
which obligations are already live and which clock is running?
That question is answerable from the same record, which is why the
MCP server exposes it directly. Point an agent at it and let it work
through provider versus deployer, Annex III classification, Article 53
duties and the national layer, with a citation to the primary source
on every claim.
Where this stands: the sources, the event record and the three MCP
tools are live. The role-by-role obligation model behind
get_obligation_for_role is being built out source by
source, and every entry ships with its citation so you can check it.
We would rather say that plainly than pretend it is finished.
Beta keys are issued by hand, usually within a day of you asking.
The feed and the health endpoint are open. Paste either line and you
have seen the product before talking to anyone.
the public wire
curl https://api.aiactradar.com/feed.xml
pipeline health, per source
curl https://api.aiactradar.com/v1/health
Case in point
On 8 July 2026, half the deadlines moved.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, amended the AI
Act itself. High-risk obligations under Annex III shifted to December
2027, the product-safety route to August 2028, while the labelling of
synthetic content moved forward. Anyone who copied the deadline table
into a wiki in June has been working from wrong dates ever since.
This is the whole argument for the product in one event. Not a feed
you read, a record you query, kept honest against the primary
sources. When something moves, the change is logged with the act that
caused it, so you can see what it was before.
There are trackers that cover every AI law on earth at a glance. This
is the opposite trade. One regulation, followed to the depth you
actually need when you ship into the EU, including the national layer
where the work really lands.
Layer one
The regulation itself
Not "Annex III applies" as a single line, but the obligations behind
it, article by article, with the trigger that switches each one on
and the evidence you are expected to hold. Amendments to the act are
tracked as they happen.
Layer two
The Commission layer
AI Office guidance, the GPAI Code of Practice, templates, delegated
and implementing acts, harmonised standards as they reach the
Official Journal. The material that decides how the text is actually
read.
Layer three
The national layer
Who supervises you, under which national act, with which sandbox and
which penalties. Germany's implementing act took effect on 29 July
2026 and made the Bundesnetzagentur the central authority. That is
the layer a global overview leaves out.
Look it up
The reference library behind the wire.
Everything the pipeline knows is also readable by humans, without a
key and without an account. These four pages answer most questions
before anyone writes us a mail.
Every event carries the official publication it came from, with the
time it was fetched and a link to the original. We aggregate and
structure, we do not interpret, and we never invent an article
reference that is not in the source. If you cannot check it, it does
not belong in the record.
EUR-Lex
EU AI Office
AI Act Service Desk
Commission Digital Strategy newsroom
European AI Board
GPAI Code of Practice
European Data Protection Board
EU Fundamental Rights Agency
Court of Justice of the EU
CEN/CENELEC JTC 21
Bundesnetzagentur
BfDI
BSI
CNIL
NL Algoritmeregister
Commission list of single points of contact
ETSI
Autoriteit Persoonsgegevens
Datenschutzkonferenz
MIRRI, Slovak government commissioner for AI
Czech Telecommunication Office
Consumer Rights Protection Centre
Two older adapters, EU Have-Your-Say and OECD.AI, are paused and
deliberately not counted above. Method and cadence per source are on
the sources page, live per-source status on the
status page.
Expect
Roughly five to ten relevant items a week.
EU regulators are not a firehose. If you want volume, this is the
wrong product. If you want to stop refreshing EUR-Lex, it is the
right one.
Expect
Manual onboarding, by a human.
Keys and endpoints are set up by hand during the beta. Slower than
self-serve, but you get direct access to the person building it.
Never
Legal advice, or a compliance verdict.
We publish structured facts with citations. What you do with them
stays your call, and your counsel's.
Pricing
Free for your team. Paid if you resell it.
Watching your own compliance footprint costs nothing, permanently.
Serving one more team is close to free for us, so we do not charge for
it. Money changes hands only when the record goes into someone else's
product.
The full reasoning, including the tier we deliberately have not built,
is on the pricing page.
Boundaries
What we are, what we are not.
We are
A structured-data publisher.
An infrastructure layer for teams that need regulatory facts in code, not in PDFs.
Bilingual EN and DE on equal footing.
Built by an engineer who reads EUR-Lex.
We are not
A law firm. We do not give legal advice.
A compliance platform. We don't run your audit trail.
A substitute for qualified counsel.
An EU institution. We aggregate; we don't certify.
AI Act Radar provides informational content only. It does not constitute
legal advice and is not a substitute for qualified legal counsel. Use of
this service does not establish any attorney-client or advisory
relationship. The content reflects publicly available regulatory and
standardisation sources at the time of ingestion and may be incomplete or
outdated.
Beta, open
Get on the wire.
Tell us what you're building and we'll set up your API key and
webhook endpoint by hand, usually within a day. Free, no payment
details, and you can walk away whenever you like.