Can your AI systems
show their sources?
From 2 August 2026, Article 50 of the EU AI Act requires providers and deployers of certain AI systems to be transparent about what is AI-generated. Regulators, auditors and your own legal team will ask a harder question next: what was this output based on?
The AI Act,
as data.
The EU AI Act is Regulation (EU) 2024/1689, CELEX number 32024R1689. Its official text is on EUR-Lex; the requests below return the full text, or one article, as JSON — with the identifiers your system should store next to every answer it grounds in the Act.
What a CELEX number or an ELI encodes, and how EUR-Lex links work: EUR-Lex URLs explained →
| Act | Regulation (EU) 2024/1689 (Artificial Intelligence Act) |
|---|---|
| CELEX | 32024R1689 |
| ELI | http://data.europa.eu/eli/reg/2024/1689/oj |
| Date of the act | 13 June 2024 |
| Official Journal | OJ L, 2024/1689, published 12 July 2024 |
| Entry into force | 1 August 2024; Article 50 applies from 2 August 2026 |
| Structure | 113 articles (art_1 … art_113) and annexes, in 24 languages |
| Official text | https://eur-lex.europa.eu/eli/reg/2024/1689/oj |
The full text, or one article, as JSON
# the whole regulation: metadata, full text, articles, annexes
curl -X POST "https://lex-api.com/api/v1/documentContent" \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber":"32024R1689"}'
# Article 50 only
curl -X POST "https://lex-api.com/api/v1/documentContent" \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber":"32024R1689","articleId":"art_50","include":["metadata","articles"]}'articleId takes art_50, 50 or Article 50; language returns any of the 24 language versions. Have an ELI or a EUR-Lex link instead? POST /api/v1/resolve turns it into the CELEX number (1 credit).
Reference an article in your app
{
"celex": "32024R1689",
"articleId": "art_50",
"eli": "http://data.europa.eu/eli/reg/2024/1689/oj",
"url": "https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689",
"corpusVersion": 1
}A record your system can keep with each answer: CELEX number and article id to fetch the provision again, ELI or EUR-Lex link for people, and the version you were served (the X-Corpus-Version response header) to show which text the answer was based on.
Four requests
that run.
Each returns a part of Regulation (EU) 2024/1689 as JSON. Set LEX_KEY to your API key.
One article
curl -X POST "https://lex-api.com/api/v1/documentContent" \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber":"32024R1689","articleId":"art_5","include":["articles"]}'Article 5, “Prohibited AI practices”, as the one item of content.articles: its id, number, title and content.
The recitals
curl -X POST "https://lex-api.com/api/v1/documentContent" \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber":"32024R1689","include":["recitals"]}'The 180 recitals as content.recitals, each with its number and text.
The annexes
curl -X POST "https://lex-api.com/api/v1/documentContent" \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber":"32024R1689","include":["annexes"]}'The 13 annexes as content.annexes, each with its title (ANNEX I to ANNEX XIII) and content.
What amends the Act
curl "https://lex-api.com/api/v1/citations/cited-by/32024R1689?citationType=amendment" \
-H "x-api-key: $LEX_KEY"The acts recorded as amending the regulation, each with its CELEX number and title. Corrigenda have their own type: ask for citationType=corrigendum.
A list of amending acts is as current as its last refresh, and an empty list is not proof that nothing amends the Act. Edge types and their sources: the citation graph →
AI output without legal grounding
is a liability.
Article 50 of Regulation (EU) 2024/1689 applies from 2 August 2026: people must be told when they interact with an AI system, synthetic content must be marked in machine-readable form, and AI-generated text published on matters of public interest must be disclosed. Non-compliance sits in the fine bracket of up to €15 million or 3% of worldwide annual turnover (Article 99(4)(g)).
LLMs answer from training snapshots. They paraphrase amended articles, cite repealed directives, and cannot tell you which version of a provision an answer was based on. Article 50(5) requires that information to be given "in a clear and distinguishable manner" — engineering-grade answers, not vibes.
In July 2026 the Digital Omnibus on AI (Regulation (EU) 2026/1744) shifted key AI Act dates — high-risk obligations moved to December 2027 and August 2028, and generative systems placed on the market before 2 August 2026 were given until 2 December 2026 for machine-readable marking. If your compliance logic hard-codes deadlines from a 2024 PDF, it was wrong within two years. What else in your stack assumes the law is static?
EU law as
infrastructure.
LexAPI gives your systems the same thing your database gives them: structured, queryable, versioned data — except the data is EU law. This is the engineering substance behind an answer that can show its legal sources, not a compliance checkbox.
25,000+ EU documents parsed into articles, sections, tables and annexes. No scraping EUR-Lex, no HTML archaeology, up to 24 languages.
Every answer your system produces can carry a CELEX number, article and paragraph — a reference a lawyer or auditor can check in seconds. A typed citation graph connects legislation and case-law in both directions.
Content-hash document versioning: store the hash alongside each AI output and you can reproduce exactly which text the answer was grounded in, months later, in an audit.
Signed webhooks on saved searches or watch lists of specific acts — when a regulation your product depends on changes, your systems find out the same day.
1. Ground a compliance assistant in the actual text
Instead of letting the model answer "what does Article 50 require?" from memory, fetch the provision and let the model quote it — with the citation attached.
curl -s https://lex-api.com/api/v1/documentContent \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{"celexNumber": "32024R1689"}' # the AI Act itselfThe response is the parsed regulation — every article addressable. Feed the relevant articles into your RAG context; emit the CELEX number, article and version hash with every answer. A document fetch served from the LexAPI corpus — like the AI Act — settles at 1 credit.
2. Know when the law changes before your users do
Subscribe a webhook to a saved search — for example, new acts from the Official Journal matching your regulatory scope — or to a watch list of the acts you depend on, and route amendments and new consolidated versions into your alerting or review queue.
curl -s https://lex-api.com/api/v1/webhooks \
-H "x-api-key: $LEX_KEY" \
-H "Content-Type: application/json" \
-d '{
"name": "AI Act watch",
"url": "https://yourapp.com/hooks/lexapi",
"searchCriteria": { "query": "artificial intelligence", "documentType": "regulation" },
"watchCelex": ["32024R1689"]
}'Deliveries are signed. This is how you catch the next Digital Omnibus the day it is published, rather than the quarter after. Webhooks are available from the Starter plan.
3. Build an audit trail your legal team can defend
For every AI output that relies on a legal source, persist three fields: the CELEX number, the article/paragraph, and the content hash of the version served. To reconstruct context around a provision — what cites it, what it cites, what amended it — pull the citation graph:
curl -s https://lex-api.com/api/v1/citations/network/32024R1689 \
-H "x-api-key: $LEX_KEY"A full citation network costs 20 credits; inbound-only or outbound-only queries are cheaper. The result: when someone asks "why did the system say this in March?", you answer with the exact text it was grounded in — not a shrug.
What LexAPI
is not.
We are engineers, so let's be precise. LexAPI does not watermark your outputs, generate disclosure banners, or certify you against the AI Act. No API can make you compliant — compliance is how you build and operate your systems. What LexAPI removes is the excuse: with citable, versioned, machine-readable EU law one HTTP call away, "we couldn't trace the legal basis" stops being an engineering limitation.
Asked &
answered.
N° 01What is the CELEX number of the EU AI Act?+
N° 02Where is the official text of the AI Act?+
N° 03How do I get the AI Act as JSON?+
N° 04How do I cite a specific AI Act article in my app?+
| Plan | Price | Credits / month |
|---|---|---|
| Free | €0 — no card | 150 (500 in the first month) |
| Starter | €15 / mo | 30,000 |
| Professional | €39 / mo | 150,000 |
| Business | €99 / mo | 500,000 |
| Scale | €349 / mo | 2,000,000 |
| Enterprise | Custom | Custom |
Document fetch: 1–5 credits (1 from our corpus) · Keyword search: 3 credits/page · Semantic search: ~5 credits · Citation network: ~20 credits. Yearly billing −20%. Full pricing →
Ship AI features that
cite their sources.
Get a free API key, fetch the AI Act, and see the parsed structure in under five minutes. 500 credits to start, then 150 a month, no card required.
Legal claims on this page: Articles 50, 99(4)(g), 111(4) and 113 of Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (Digital Omnibus on AI, in force 27 July 2026). Verified November 2026 — this page describes engineering capabilities, not legal advice.