PageMax

All articles

EU RULES · AI TRANSLATION

Does an AI translation need its own review under the AI Act?

Whether a lawyer’s approval of an English page carries the AI Act’s exception into its machine translation.

Grey shapes standing on a fine line with their rippled reflection below, one reflected circle in green and an exact match for its original: the translation a person has checked
Listen to this article

0:00--:--

A lawyer stands behind an English page on personal injury time limits. They wrote it themselves, or they checked and corrected a draft that AI wrote, and the firm publishes it in English. A machine then translates it into Polish, and the Polish page goes live without anyone reading it.

Either way, the English page needs no AI label under Article 50(4) of the EU AI Act: a person wrote it, or a person reviewed it and the firm holds editorial responsibility for it. The question is the Polish page. Does it need an AI label, or its own review? PageMax translates law firm websites, so we had to answer that.

Key facts

Topic Key fact
The rule Article 50(4), second subparagraph, of the EU AI Act, applying since 2 August 2026 (Article 113; para 153).
Who must act The deployer: on our reading, the firm that decides to use machine translation for its pages (paras 12 and 14).
Which pages Published text informing the public on matters of public interest, including the administration of justice (para 131). On our reading, not menus or buttons; usually not advertising (the examples after para 131).
The exception Human review or editorial control, and a person holding editorial responsibility: "two cumulative conditions" (para 133).
A machine translation On our reading, new AI-generated text that the English review does not cover (para 136; the examples after para 138).
If no one reviews it A clear label at the start of each translated page (Article 50(5); the examples after para 143).
Digital Omnibus No delay for Article 50(4). Only providers' marking and detection duties under Article 50(2), for systems on the market before 2 August 2026, moved to 2 December 2026 (Article 111(4)).
The guidelines Content approved on 20 July 2026. They apply only from formal adoption, of which we found no record as of 15 October 2026 (C(2026) 5054).
Fines Up to EUR 15 million or 3% of worldwide turnover, whichever is higher; whichever is lower for SMEs and small mid-caps (Article 99(4)(g), (6) and (6a)).
Sources checked on 15 October 2026.

What a law firm should do

On our reading, a firm that machine-translates its website should:

  1. List the pages a machine translates, and separate those that explain the law from menus, buttons and pages that only advertise the firm (para 131).
  2. Decide, for each page that explains the law, before it goes live: a person reviews the translation against the approved English, or the page carries an AI label at the top (Article 50(4) and (5)).
  3. Choose a reviewer for each language who reads both languages well and knows the legal subject well enough to spot a shifted term, sending any doubt to the lawyer (para 134).
  4. Name who holds editorial responsibility, with contact details a reader of the translated page can find (para 138).
  5. Send a translation back for review whenever the English changes (para 136).
  6. Check translations published since 2 August 2026, including ones produced before that date (para 154).
  7. Write the approach down in a short review policy (Code of Practice, Section 2, Commitment 4). It need not log each review, but every published translation needs its review or its label (para 143).

Do machine-translated pages need an AI label?

Only some do. Article 50(4) covers AI-generated text published to inform the public on matters of public interest. The Commission's guidelines on Article 50 say that "short texts which do not materially communicate knowledge, opinions or facts, cannot be deemed to inform the public" (para 131), which on our reading puts menus and buttons outside the rule. Their examples of text outside the rule include text that is part of "a company's advertisement or product descriptions (not including any claims related to e.g. health, consumer safety or sustainability)" (the examples after para 131). For most translated websites, a machine translation needs no visible AI label.

A law firm's pages that explain the law to the public are different. The same paragraph names "the administration of justice" as a matter of public interest. Irish rules on legal advertising point the same way: S.I. 644/2020 treats "an article on a legal topic written by a legal practitioner" as information on the law rather than an advertisement, where no part of its space is paid for by or on behalf of the practitioner (regs 2 and 10(a)(ii)). Whether an AI draft that a lawyer approved, or a machine translation of one, counts as "written by a legal practitioner" is not established.

We looked at the public interest question in our earlier article on Article 50, and here we assume a page that explains the law is in scope. For those pages, the label question turns on whether the lawyer's review of the English also covers the translation. If it does not, the translated page needs its own review or an AI label.

The Article 50(4) exception, human review and editorial responsibility

The second subparagraph of Article 50(4) reads:

"Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated. This obligation shall not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences or where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content."

Recital 134 states the exception in nearly identical words. The wording of Article 50(4) is unchanged in the consolidated text of 27 July 2026, after the amendments made by Regulation (EU) 2026/1744 (the Digital Omnibus on AI), which, within Article 50, changed only paragraph 7, on codes of practice. Its only transition for Article 50 is for providers' marking and detection duties under Article 50(2) (Article 111(4)). The deployer's duty in Article 50(4) has applied since 2 August 2026 (Article 113; para 153).

The Commission's guidelines read the exception as "two cumulative conditions": human review or editorial control, and a person holding editorial responsibility (para 133). Human review means "the deliberate examination of the substance of the content by one or more natural persons possessing relevant knowledge and professional judgement pertaining to the subject matter under scrutiny", and "Fact-checking the accuracy of the content is a minimum requirement" (para 134). Spell-checks, "the mere existence of an editorial policy, automated review processes or cursory editorial approval" do not count (para 135).

Editorial responsibility, the second condition, is the easier half. The person holding it must have "the ultimate legal responsibility over the publication of the content, including the human review or editorial control", and their identity and contact details "should be made publicly available on an easily findable location" (para 138). A firm that holds that responsibility for its English pages can hold it for their translations, provided a reader of the Polish page can find who that is.

Is a machine translation AI-generated text?

Article 50(4) applies to a deployer of an AI system that "generates or manipulates text". The guidelines describe generation as producing synthetic material, and manipulation as altering "already existing content (whether synthetic or not)" (paras 58 and 59). They also list "automated translation or transcription tools" as AI systems, in their examples of systems that do not interact directly with people (the examples after para 31). A machine translation takes existing text and returns new text in another language. On either description, the output is AI-generated or manipulated.

The clearest evidence of the Commission's view is an example. Among the texts that qualify for the human review exception, the final guidelines list "AI-supported translation of a human-written article whereby the translation has undergone human review" (the examples after para 138). That example only works if an AI translation falls under Article 50(4) to begin with. It also puts the review on the translation, even though the source in that example was written by a person and needed no review at all. On our reading, a translation of an AI-drafted article that a lawyer approved cannot need less.

The standard editing exception for translation is in Article 50(2), not 50(4)

The argument the other way starts in Article 50(2). That paragraph requires providers of generative AI systems to mark their outputs in a machine-readable format. It "shall not apply to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof". Recital 133 presents this as a limit on "this marking obligation".

The Commission changed its mind about translation in this section. The May 2026 consultation draft listed "AI-generated translations and summaries of text" among the "semantic changes that require marking" (draft para 86). The final guidelines moved "AI-generated translations of text" into the list of standard editing and minor alterations that benefit from the exception, in the same bullet as grammar correction and spellchecking (the examples after para 92). Summaries stayed on the other side.

This does not answer our question, because the two paragraphs are different rules for different people:

Point Article 50(2) Article 50(4), second subparagraph
Who it binds Providers of AI systems that generate synthetic content Deployers publishing AI-generated or manipulated text to inform the public on matters of public interest
What it asks Machine-readable marking of outputs Telling readers the text is AI-generated or manipulated
Its exceptions Standard editing, or no substantial change to the input or its meaning; law enforcement Human review or editorial control, with editorial responsibility; law enforcement
Translation Standard editing in the final guidelines (the examples after para 92) No standard editing exception; the one translation example puts the review on the translation (the examples after para 138)
Digital Omnibus Until 2 December 2026 for systems on the market before 2 August 2026 (Article 111(4)) None: applies from 2 August 2026 (para 153)

The Article 50(4) labelling duties "apply in addition" to the Article 50(2) marking duties (para 111), so an exception written into one is not written into the other.

Nor does either limb of the exception fit translation well, and the guidelines do not say which one applies. Standard editing "does not involve generating new content" (para 90), yet a translation replaces every word of its input. The closer fit is the second limb, a system that keeps "the semantics" of its input. But whether a translation kept the meaning is exactly what a reviewer is there to check. The provider's exception assumes the meaning survived. The deployer's exception asks a person to confirm it.

There is a counter-argument. The guidelines' Article 50(4) chapter borrows the Article 50(2) chapter's account of generation and manipulation (paras 113 and 130), which puts content within Article 50(2) where it is AI-generated or manipulated "beyond standard editing" (para 59). So, the argument runs, a translation is not "manipulated" text for Article 50(4) at all. The chain is loose (para 113 sits in the deep fake section, and para 130 does not point to para 59), and the translation example after para 138 would serve no purpose if translations were outside Article 50(4). But no authority has settled the point.

Does a lawyer's review of the English cover the translation?

The strongest case that it does runs like this. The lawyer checked the substance in English, and a faithful translation adds no facts. The Commission itself lists translation among "standard editing and minor alterations" exempt from marking (the examples after para 92). The firm holds editorial responsibility for the whole site. And para 136 voids the exception for "substantive" AI intervention after sign-off, which a faithful translation arguably is not.

We still think the answer is no, for four reasons.

  1. The wording of the exception. It applies where "the AI-generated content has undergone a process of human review". On the Polish page, the AI-generated content is the Polish text, and no person examined it.
  2. The test for review. Para 134 asks for "deliberate examination of the substance of the content", with fact-checking as the minimum. A lawyer who approved the English examined a different text. A lawyer who does not read Polish cannot examine the Polish one.
  3. AI changes after sign-off. Para 136 says: "Where AI systems are used to modify, supplement, or reformulate content following editorial sign-off, the resulting content must be treated as AI-generated or manipulated for the purposes of Article 50(4) AI Act. Any substantive AI intervention occurring after the human review or editorial control process has taken place will therefore cause the exception to become void." A translation reformulates every sentence after sign-off, and every word the reader sees is the machine's. On our reading that is substantive intervention, whatever the guidelines say about translation for marking. Using the marking examples to switch off the deployer's duty would let an exception the Act confines to "this obligation" in Article 50(2) do work in Article 50(4).
  4. The translation example. The example after para 138 attaches the review to the translation, not to the source.

The purpose of the rule points the same way. The transparency duties exist "to reduce the risks of impersonation, deception, misinformation, manipulation at scale, and fraud" (para 9). A reader of the Polish page relies on the Polish words alone. If nobody read those words before they were published, telling the reader so is what the Act asks for.

One practical point follows. AI text published before 2 August 2026 needs no label after the event, but text generated before that date and published on or after it does (para 154). A translation made today of an older article is new AI-generated text, whatever the date of the English.

What human review of a translation must include

Para 134 allows "one or more natural persons", so the person who reviews the translation does not have to be the lawyer who approved the English. What the reviewer needs is "relevant knowledge and professional judgement pertaining to the subject matter under scrutiny".

ISO 17100, the international standard for translation services, has a second qualified person, the reviser, check a translation against its source before it is delivered (clause 5.3.3), and ISO 18587 sets out how a person post-edits machine translation output so that the result is comparable to a human translation. A firm using a machine has no reason to expect less than that second pair of eyes.

For a translation of an approved page, we read the subject matter under scrutiny as whether the translation says what the approved English says. The law was checked in the English review. The fact-check for the translation is a check against that approved text, line by line: the same time limit, the same conditions, the same exceptions, nothing added and nothing lost.

That takes more than fluency. The reviewer needs a strong command of both languages and enough understanding of the legal subject to notice when a legal term has shifted, and refers anything uncertain to the lawyer. A specialist legal translator, or a member of staff who works with the lawyer in that area of law, can meet that; a lawyer who reads the language is the safest choice. Whether a translator who is not a lawyer has "relevant knowledge" in the guidelines' sense is not established.

On our reading, these do not count as review of a translation:

  • The approved English alone. It is the source, not the text the reader sees.
  • Reading the translation on its own. Without the approved English beside it, a reviewer checks the Polish against their own view of the law, not against what the lawyer approved, and can miss a change that still reads as plausible law. On our reading, para 134's fact-check for a translation is a check against the approved text.
  • A glossary or translation memory. These make errors less likely and are worth having, but they are tools, and para 135 rules out "automated review processes".
  • A second machine checking the first. The guidelines' failing examples include "AI-generated articles that are reviewed and edited by another AI system" where a human does only a superficial check (the examples after para 138).
  • Reviewing only what a tool flags. If a tool sends a person only the sentences it doubts, the rest are published on an automated review process alone (para 135).
  • Reviewing a sample. The exception attaches to each text that "has undergone" review, and the duty applies to "each output" (para 143), so checking one page in ten leaves nine pages unreviewed.

Editorial control, the alternative to human review, does not change this. Para 134 describes control "exercised in practice" by someone with "the authority to approve, alter or reject the substance of the text based on substantive grounds". A lawyer who cannot read the Polish text has that authority but cannot exercise it on that text.

A firm can still write its approach down. Para 137 points deployers that are not media businesses to a code of practice assessed as adequate: the Code of Practice on Transparency of AI-generated Content (final version, 10 June 2026), whose Section 2 the Commission found adequate for Article 50(4) and (5) in an opinion of 8 July 2026 (paras 38 and 40).

For a signatory that is not a media business, Commitment 4 in that Section asks for a policy for human review before publication. A translation policy built on it and on para 138 would, on our reading, name the person with editorial responsibility (name, role and contact details) and who reviews each language, and say that every translated page is checked against the approved English before it is published, or else labelled, and goes back for review when the English changes (para 136).

The policy "does not entail having to document individual instances of human review", though signatories "may record additional information on the nature of the review" (the Code, Section 2, Commitment 4). A firm that has not signed the Code should expect more questions: deployers outside it "may also be subject to such requests with regard to their labelling practices" (para 148). That is reason enough to record who checked each translated page, and when.

Where the AI label goes on a translated page

If the exception does not apply, the label must be given "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure", and it must "conform to the applicable accessibility requirements" (Article 50(5)). The guidelines' example of first exposure for text is "at the start of an AI-generated or manipulated text publication on matters of public interest" (the examples after para 143).

The guidelines accept "labelling or disclosure methods (including disclaimers)" that are "clear and perceivable" (para 132), but not information that "can be easily overlooked or missed by natural persons", such as "part of terms of use that are often not read by users" (para 142). On our reading, a note in the footer or in the site's terms does not label a translated page. And the duty "applies to each output of an AI system with respect to any natural person exposed to the content" (para 143), so each translated page carries its own label, for every reader who opens it.

The Act does not say which language the label must be in. The guidelines ask for information that is "easy to understand by and accessible for the natural person concerned" (para 142), so on our reading the label belongs in the language of the page: a Polish reader is told in Polish.

The Commission has published EU icons for labelling: "The use of these EU icons is optional, but the labelling requirements under Article 50 AI Act are not" (European Commission, EU icons). Firms that sign the Code commit to an icon or label built on the capitalised acronym "AI", in English unless national language laws rule that out (Section 2, Measure 1.1(a)), placed, for example, "above or at the top of the text, near the headline of the text, or in the colophon at the beginning of the text" (Sub-measure 1.2.2(f)). On our reading, a short line in the page's language can sit beside it.

Who is the deployer when an agency or website provider translates?

Article 50(4) puts the duty on the deployer, the person under whose authority the AI system is used. The guidelines read that authority as "assuming responsibility over the decision to deploy the system and over the manner of the actual use of the system (including its outputs)", which "does not necessarily require technical control" (para 12). A legal person "remains a deployer even if it involves third parties (e.g. contractors, freelancers) in the operation of the system on its behalf and under its responsibility and control" (para 14).

The guidelines also mark the limits. A company that "merely commissions an advertising agency to produce an advertisement, without taking decisions and exercising control over whether and how the advertising agency uses AI in the production process, is not a deployer" (the examples after para 14). Nor is a host whose role "is limited to disseminating or transmitting AI-generated or manipulated content created by third parties" (para 16).

On our reading, a firm that chooses machine translation, or tells its agency or website provider to use it, decides whether and how AI is used, and is the deployer for its translated pages. Its contract with the provider should then say who reviews each translation and who adds the label.

Why it matters on an Irish law firm's website

In Census 2022, 751,507 people in Ireland said they spoke a language other than English or Irish at home, Polish most often (123,968), and 95,170 of them said they could not speak English well or at all (CSO, Census 2022 Profile 5; table F5004). For those readers, the translation is the only version of the page they can rely on.

Take the time limit page. For most personal injury claims the limit is two years from the date the cause of action accrued, or from the injured person's "date of knowledge" if that is later (Statute of Limitations (Amendment) Act 1991, s.3(1), as amended by the Civil Liability and Courts Act 2004, s.7). The date of knowledge is the date the person first had knowledge of certain facts, and that includes knowledge they might reasonably have been expected to acquire (s.2(1) and (2)). The facts include:

  • that they were injured;
  • that the injury was significant;
  • that it was attributable, in whole or in part, to the act or omission complained of;
  • who the defendant is.

The Act adds that "knowledge that any acts or omissions did or did not, as a matter of law, involve negligence, nuisance or breach of duty is irrelevant" (s.2(1)).

Now suppose the translation renders "date of knowledge" as words meaning "the date you found out you could make a claim". It reads naturally, but it says the opposite of s.2(1), and a reader who waits for that date can lose the claim. No one reading only the English would see the error. A glossary would catch it only if it held that exact phrase. And the lawyer who approved the English never saw the Polish.

Who enforces Article 50 against a law firm in Ireland is not settled. S.I. 366/2025, as amended by S.I. 405/2026, makes Coimisiún na Meán the market surveillance authority for Article 50 only where the provider or deployer is an intermediary, audiovisual media, sound broadcasting, relevant designated online or hosting service (reg 4(3)). As of 15 October 2026 we have found no designation that expressly covers a firm publishing on its own website, and the entries in s.78 of the Regulation of Artificial Intelligence Act 2026 that would bring Coimisiún na Meán and the Data Protection Commission into its adjudication and fines procedure (paras (a) and (e) of the definition of "applicable market surveillance authority") are shown as not yet commenced (Irish Statute Book, updated to 6 October 2026).

The maximum fine for breaching Article 50 is EUR 15 million or, for an undertaking, 3% of worldwide annual turnover, whichever is higher (Article 99(4)(g)); for SMEs and small mid-cap enterprises it is whichever is lower (Article 99(6) and (6a)). For a law firm the larger risk is a reader who acts on a page that says something no lawyer approved.

The Law Society's guidelines on generative AI (December 2025) list translation among the uses of these tools and leave the AI Act's disclosure duties "beyond the scope" of the guidance. The guidance does not deal with website content, but its statement on work "submitted externally to clients and the courts" points the same way: "The duty to ensure accuracy and reliability remains firmly with the solicitor."

What we built, and why

In PageMax, no article drafted with AI for a law firm's website goes live until someone at the firm approves it. Where a firm already has its own translation of a page, that translation comes first. Where machine translation fills a gap, this is how we have built it, and it is new:

  • New machine translations wait. With review on, which is the default, a machine-translated page is not published until a person at the firm has gone through it beside the approved English, paragraph by paragraph, the title included, and confirmed that they read the language, understand the legal subject well enough to check it, and referred any doubt to the lawyer. That person can be a member of staff who translates, with a login only for this.
  • Once checked, the AI line comes off. The page carries no AI line, because on our reading it now meets the exception.
  • If the English changes, the translation goes back to waiting. A check against an earlier version of the English is not a check of the current one.
  • A firm can decide not to review. In that case new translations go live without waiting for a check, with a line at the top of the page, in the page's own language, saying it is an AI translation of the English original. So the reader learns that a machine wrote these words from an English original.

We assume the firm holds editorial responsibility for its pages. Review is on by default, and the firm's admin can turn it off. What we would not build is a third option for a page's text: an unreviewed translation with no label, resting on the English approval. Menus and buttons are translated automatically: on our reading they are short texts that do not inform the public (para 131). The other words every page shares, such as the footer, are translated automatically too, and are not held for review.

We did not require a lawyer to review every translation: a legal translator, or staff who work with the lawyer in that area of law, can meet the reviewer test above. And for many readers, the alternative to a reviewed translation is their browser's own machine translation, which nobody at the firm has read.

What is still open

None of these is settled law:

  1. Is a page that explains the law "public interest" text, and what of a page that both informs readers and wins clients (para 131)?
  2. Is translation "generate or manipulate" for Article 50(4) at all (para 59), and is it "substantive AI intervention" under para 136?
  3. How do the guidelines treat a machine translation of an AI draft that a lawyer approved? Their example after para 138 has a human-written source.
  4. How much legal knowledge does a reviewer who is not a lawyer need, and when must the lawyer be involved?
  5. Who is the deployer when a website provider drafts and translates and the firm approves (paras 12, 14 and 15)?
  6. Which Irish authority supervises Article 50 for a firm publishing on its own website (S.I. 366/2025, reg 4, as amended)?

The Code of Practice helps with review but not with translation. It mentions translation once, in the providers' section, where "translation cycles" are among the processing operations a marking solution should survive (Section 1, Measure 3.3). It expects a machine's marks to survive translation, and says nothing about who reviews one.

The guidelines themselves are non-binding, and call themselves "a first interpretation", which the Commission may amend (paras 5 and 155). On the Commission's own terms they may not apply yet, and the reason is translation. On 20 July 2026 the Commission approved their content, citing "the time necessary for the translation of the guidelines into all languages". It said they "will be formally adopted by the Commission at a later date, when all language versions are available. It is only from that moment that these Guidelines will be applicable" (C(2026) 5054 final).

As of 15 October 2026 we have found no record of that formal adoption. The Commission's library page says "The Commission adopted these guidelines", but it still offers only the English text and the Communication.

Review that happened

Article 50(4) is a rule about where a person has to stand when AI is used to write. With a translation, the English was reviewed and the page looks finished, but the text a Polish reader acts on was never read. A label that says so keeps faith with that reader, and so does a review that actually took place.

Frequently asked questions

Do I need to label a machine translation under the AI Act?

On our reading, yes, if the page explains the law to the public and nobody has reviewed the translation (Article 50(4); para 131). No label is needed once someone with the language skills and enough knowledge of the legal subject has reviewed it, and someone holds editorial responsibility for it. Shop, product and marketing pages are usually outside the rule.

Does the AI Act exempt machine translation from AI labelling?

No. The Commission's guidelines list translation as standard editing, but that exception is in Article 50(2) and covers only the machine-readable marking that AI providers add to their output. The firm that publishes the text has its own duty under Article 50(4), with its own exception: human review or editorial control, with someone holding editorial responsibility.

Where does the AI label go on a translated page?

At the start of the translated text, on every translated page, in words the reader understands (Article 50(5); paras 142 and 143). A line in the terms of use, or anywhere else a reader can easily miss, does not count (para 142). The EU icons are optional; the label is not.

Does a lawyer's review of the English page cover the translation?

On our reading, no. The translation is new AI-generated text, and the Commission's own example of the exception is a translation that has itself undergone human review. Approving the English does not check whether the translation says the same thing.

Who can review a legal translation?

Someone with a strong command of both languages and enough understanding of the legal subject to notice when a legal term has shifted, who refers anything uncertain to the lawyer. How much legal knowledge that takes for a reviewer who is not a lawyer is not settled.

Can a firm review only some translations, or only those a tool flags?

Only if every page it does not review carries the label, because the duty applies to each output (para 143). A tool that flags likely errors can help a reviewer, but automated review processes do not count as review (para 135).

Did the Digital Omnibus delay Article 50 for translations?

No. Regulation (EU) 2026/1744 gave only providers' machine-readable marking and detection duties under Article 50(2) a transition to 2 December 2026, for systems on the market before 2 August 2026 (Article 111(4)). The deployer's duty to label text under Article 50(4) has applied since 2 August 2026 (para 153).

Does Article 50 apply to a firm outside the EU?

It can. The AI Act reaches deployers outside the EU "where the output produced by the AI system is used in the Union" (Article 2(1)(c)). The guidelines apply that where the deployer "foresees dissemination and use of the AI outputs in the Union", and their example is posting deep fakes "on the globally accessible internet". A deployer is not bound where its content reaches EU audiences "through channels that are unforeseeable and outside their control" (para 13). How that applies to a firm outside the EU whose translated pages readers in the EU can open is not settled.

Have the Commission's Article 50 guidelines been adopted?

Not formally, as far as we can find. The Commission approved their content on 20 July 2026 and said they apply from their formal adoption, once every language version is available. As of 15 October 2026 we have found no record of that adoption. Article 50 itself has applied since 2 August 2026 either way.

This is not legal advice. We build the system, we do not advise on the law, and no one should act on a vendor's reading of a regulation. If Article 50 matters to how your firm publishes, take your own advice on it.

Sources and further reading

Be the law firm cited in ChatGPT

We're opening PageMax to a small number of firms. Book a demo.