06-reference/research

dsa political ads emfa ai labelling duties

2026-11-01·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
ai-actdsaemfapolitical-advertisingsanity-checkcompliance

No. All three miss us, and two of the three impose no AI-content labelling duty on anybody. The parent verdict stands unchanged.

Not legal advice. This is an operational read of primary text by a non-lawyer. A real EU media lawyer should see this before it becomes commercial policy, and the place that matters most is the European Media Freedom Act "press publication" question in follow-up 2.

Emphasis added in every quoted passage below is mine, not the source's.

The question

Verbatim: "Do the DSA, the Political Advertising Regulation, or the European Media Freedom Act impose AI-content labelling duties that the AI Act Article 50(4) editorial carve-out does not reach?"

Context: [[2026-09-03-ai-act-article-50-transparency-guidelines]] produced the only concrete change to a shipping RDCO surface in the whole AI Act thread. Sanity Check needs a publicly findable editorial-responsibility statement, naming a person or function with contact details, to hold the Article 50(4) carve-out. That verdict rests on the carve-out being the only labelling duty in play. This brief tests the three instruments that could break that assumption: the Digital Services Act (Regulation (EU) 2022/2065, "DSA"), the Political Advertising Regulation (Regulation (EU) 2024/900), and the European Media Freedom Act (Regulation (EU) 2024/1083, "EMFA"). It also reads the Commission Article 50 guidelines sections on interplay with other Union acts that the parent pass skipped.

Answer, up front

Instrument AI-content labelling duty? Duty-bearer Editorial exemption? Does it reach RDCO?
DSA (Reg. (EU) 2022/2065) No. Art. 35(1)(k) is a listed example of a risk-mitigation measure, not a standalone labelling rule, and on its own wording it covers image, audio and video but not text Providers of very large online platforms (VLOPs) and very large online search engines (VLOSEs), per Art. 35(1) None, and none needed: Art. 2(2) excludes any service that is not an intermediary service No. Own-content publishing is not an intermediary service at all; separately, VLOP designation needs 45 million monthly EU recipients
Political Advertising Reg. (Reg. (EU) 2024/900) No. The only two artificial-intelligence references in the whole Regulation are in Art. 19, and they are about AI used to target or deliver an ad, not AI used to create content Political advertising publishers, sponsors, providers of political advertising services Yes, Art. 1(2), and broader than Art. 50(4) in one respect: no publication-of-identity condition is attached Not applicable. RDCO runs no political advertising. One forward trip-wire, below
EMFA (Reg. (EU) 2024/1083) No. Zero content-labelling provisions. Art. 18(1)(e) is the inverse: a declaration that you do not publish AI content without human review Art. 18 duty falls on VLOP providers; Art. 6(1) falls on media service providers but is service-level transparency, not content labelling The concept runs the other way: EMFA Art. 2(8) is one of the sources the Commission tells us to read Art. 50(4) against No labelling duty. Whether RDCO is a "media service provider" owing Art. 6(1) is genuinely open, and that duty is about ownership and contact details, not AI

Verdict on the parent brief: STANDS, unchanged. No instrument examined here adds an obligation. The parent's single shipping-surface action item, the ¶138 editorial-responsibility statement, is still the right one and still the only one on a surface. The EMFA does not add a second duty; what it offers is a possible reuse of the same document, conditional on two things RDCO probably cannot satisfy (below). I am not manufacturing a new action item to make this brief feel productive.

What we already know (from the vault)

What the web says

Convergences and contradictions

Instrument by instrument, on the operative text

DSA: wrong architecture, not just wrong threshold

Two independent walls, and the first matters more than the headline threshold.

Wall one, Article 2(2): "This Regulation shall not apply to any service that is not an intermediary service or to any requirements imposed in respect of such a service, irrespective of whether the service is provided through the use of an intermediary service." Intermediary service is closed-listed in Art. 3(g) as mere conduit, caching, or hosting, where hosting is "the storage of information provided by, and at the request of, a recipient of the service." Sanity Check and the rdco.dev sites publish RDCO's own content at RDCO's own request. Recital 18 confirms the direction: the liability exemptions are unavailable for "information provided not by the recipient of the service but by the provider of the intermediary service itself, including where the information has been developed under the editorial responsibility of that provider." Own editorial content sits outside the framework rather than being exempted within it. That is why the DSA has no Art. 50(4)-style editorial carve-out; it never needed one.

Wall two, the scale thresholds, which only matter if wall one somehow fell: Art. 35(1) binds "providers of very large online platforms and of very large online search engines," and Art. 33(1) sets that at 45 million average monthly active recipients in the Union plus a Commission designation decision under Art. 33(4). Art. 19 separately excludes micro and small enterprises from the whole online-platform section. A one-person newsletter is not within several orders of magnitude.

And the duty itself is weaker than its reputation. Art. 35(1): providers "shall put in place reasonable, proportionate and effective mitigation measures, tailored to the specific systemic risks identified pursuant to Article 34 ... Such measures may include." Point (k) is item eleven on a menu. Its material scope differs from Art. 50(4) in two directions. Guidelines ¶126 names the two distinctions it sees: a broader material scope, because point (k) catches content manipulated "regardless of the technologies used," and a different personal scope, VLOP and VLOSE providers rather than deployers. A third difference comes from the wording of point (k) itself rather than from ¶126: it reaches only "a generated or manipulated image, audio or video," so AI-generated text is outside it. So the reading that Art. 35(1)(k) is a general labelling duty is not supported by the text, for three separate reasons: it is illustrative, it is platform-bound, and on its face it does not cover text.

One genuinely useful thing in ¶126 points the other way. Where a VLOP provides labelling tools, "those deployers can rely on such tools to fulfil their transparency obligation under that provision within the context of the VLOP or VLOSE used." If RDCO ever posts content that does clear the deep-fake bar to LinkedIn or X, the platform's own AI-content label can discharge the Art. 50(4) first-subparagraph duty for that post.

Political Advertising Regulation: inapplicable, with one sponsorship trip-wire

Crisply: no AI-content labelling duty exists in this Regulation, and the Regulation does not apply to RDCO because RDCO runs no political advertising.

The labelling duty that does exist, Art. 11(1), requires each political advertisement to carry a statement that it is a political advertisement, the sponsor's identity, the linked election or process, a statement if targeting was used, and a transparency notice under Art. 12. None of it concerns AI origin. The only two artificial-intelligence references in the instrument are in Art. 19, which requires transparency about "whether an artificial intelligence system has been used to target or deliver the political advertisement." That is a duty about the targeting pipeline, not about who or what wrote the words.

The editorial carve-out, Art. 1(2), is worth noting because it is broader than Art. 50(4) in the one way that matters to us: "Political opinions and other editorial content, regardless of the medium through which they are expressed, that are subject to editorial responsibility shall not be considered to be political advertising unless specific payment or other remuneration is provided for, or in connection with, their preparation, placement, promotion, publication, delivery or dissemination by third parties." No publication-of-identity condition, no fact-checking minimum, no post-sign-off voiding rule. Art. 1(3) adds that political opinions expressed in a personal capacity are out.

The one forward trip-wire. The Sanity Check monetization plan in [[revival-strategy]] is third-party sponsorship, with outreach opening at 1,000 subscribers. Art. 1(2)'s carve-out falls away where a third party provides specific payment for or in connection with the editorial content. Combine that with Art. 3(2)(b), which defines political advertising to include a message "liable and designed to influence ... a legislative or regulatory process, at Union, national, regional or local level," and a paid-sponsored Sanity Check issue advocating a position on AI regulation could land inside Art. 11, if the sponsor's payment is tied to that content rather than being a generic per-issue slot. The distinction between a generic sponsor slot and payment for a position is exactly where Art. 1(2) turns, and it is not resolved by the text alone. Worth a lawyer's eye before the first sponsored issue, not before the first subscriber.

EMFA: not a duty, and a reuse that is probably out of reach

The EMFA contains no content-labelling provision of any kind. Its only artificial-intelligence provision is Art. 18(1)(e), and reading it carefully inverts the worry:

"Providers of very large online platforms shall provide a functionality allowing recipients of their services to: ... (e) declare that they do not provide content generated by artificial intelligence systems without subjecting it to human review or editorial control."

The obligation is on the platform to offer the box. The publisher's side is a self-declaration whose payoff is the Art. 18(4) procedural protections: a statement of reasons and a 24-hour window to reply before a VLOP suspends or restricts visibility of your content. It is a declaration that you do keep a human in the loop, which is the opposite of a label, and the consequence of not making it is losing a protection rather than incurring a duty.

Art. 6(1) is the one real duty the EMFA places on publishers, and it is service-level, not content-level: legal name and contact details, influential owners, beneficial owners, and the annual totals of public funds received for state advertising and of advertising revenue from third-country public authorities or entities, all "easily and directly accessible to the recipients of their services." Art. 6(3) adds, for providers of news and current-affairs content, measures to guarantee editorial independence and to disclose conflicts of interest. None of this touches AI origin, and none of it asks for a named function holding editorial responsibility. Keeping the service-level and content-level duties apart is the whole reason this instrument reads as alarming and resolves as harmless.

Does RDCO fall inside the duty-bearer class? Honestly uncertain. EMFA Art. 2(1) defines a media service as one whose principal purpose is "providing programmes or press publications, under the editorial responsibility of a media service provider, to the general public." Art. 2(5) imports "press publication" from Art. 2(4) of the Digital Single Market copyright directive, Directive (EU) 2019/790: a collection mainly of literary works of a journalistic nature, constituting an individual item in a periodical or regularly updated publication under a single title, informing the general public, "published in any media under the initiative, editorial responsibility and control of a service provider." Recital 56 of that directive expressly includes news websites and "subscription-based magazines," and excludes blogs only where the activity is not carried out under a service provider's initiative, editorial responsibility and control. Ray Data Co LLC supplies exactly that initiative and control over Sanity Check, so the blog exclusion does not obviously rescue us.

The weak link is Art. 2(2)'s "whose professional activity is to provide a media service." The Article 18 guidelines stress that recital 9 of the EMFA "highlights the professional character of the activity as a determining factor," while confirming the definition covers "a wide spectrum of professional media actors, including freelancers." Sanity Check is free, RDCO is dormant, a media service is a service within the meaning of Articles 56 and 57 of the Treaty on the Functioning of the European Union (normally provided for remuneration), the EMFA has no territorial-scope article, and enforcement runs through national regulatory authorities over providers within Member State jurisdiction. So: arguable in principle, no practical duty, and in any case not a labelling duty. Filed as follow-up 2 rather than resolved here.

The Article 18 reuse, and why it is probably out of reach. Section 4.4 of the Article 18 guidelines describes the optional information a media service provider may share alongside the point (e) declaration as relating to "specific functions or roles within a media service provider's organisation that are entrusted with editorially reviewing content generated by artificial intelligence systems, or with taking editorial responsibility for such content" (¶40). That is close to the artefact AI Act guidelines ¶138 asks for, described in similar words by the same institution five months earlier. The temptation is to call that a second reason to write the statement. Two features of the text block it:

  1. The declaration is cumulative, not a menu. Guidelines ¶1: "To benefit from these safeguards, media service providers need to declare that they fulfil a number of elements, as listed in Article 18(1), first subparagraph, points (a) to (g)." Point (a) is a declaration that you are a media service provider, which is the open question above. Point (d) requires declaring that you are "subject to regulatory requirements for the exercise of editorial responsibility in one or more Member States and to oversight by a competent national regulatory authority or body," or that you adhere to a co-regulatory or self-regulatory mechanism "widely recognised by and accepted in the relevant media sector in one or more Member States." A US-established one-person newsletter has no evident route to point (d).
  2. The ¶40 information is expressly optional. Guidelines ¶17 makes the whole supporting-information feature "strictly subject to media service providers' sole discretion," and says its absence "should not render the declaration incomplete."

So the honest statement is narrower than it first looked: the EMFA asks RDCO for nothing, and the convergence of wording is a drafting convenience available if RDCO ever becomes a media service provider with a point (d) route. It is not a second duty, and it is not a second reason the statement must exist.

Synthesis for RDCO

The answer to the question as asked is no, three times, and the reasons are different enough to keep separate. The DSA misses us on architecture: Art. 2(2) says the Regulation does not apply to services that are not intermediary services, and a newsletter publishing its own content is not one. The Political Advertising Regulation misses us on subject matter: it contains no AI-content labelling duty for anyone, and we run no political advertising. The EMFA misses us because it has no labelling duty to miss anyone with. The parent brief's verdict therefore stands without modification. The single shipping-surface action item is unchanged, it is still open as far as vault records go, and its two companion policy items from the parent brief are untouched by anything here.

What this pass did buy is a tested negative, which is worth more than it reads. The parent verdict was a single-instrument conclusion: we owe a label unless we hold a carve-out, and the carve-out has a form requirement. A single-instrument conclusion in EU digital law is exactly the kind of thing that breaks when the second and third instruments arrive, because the usual pattern is overlapping obligations with mismatched exemptions. That pattern does not appear here. All three candidate instruments route their AI-adjacent provisions to platforms, and the only one that places a real duty on publishers, EMFA Art. 6(1), places a service-level ownership-and-contact duty that has nothing to do with AI origin. The content-versus-service distinction is the thing to keep: these are different obligations, they live in different instruments, and collapsing them is how a small publisher talks itself into compliance work it does not owe.

The thread's recurring lesson repeats. "Who are you" resolved the question twice more here: Art. 35(1)(k) DSA misses us because we are a publisher rather than a platform, and EMFA Art. 18 misses us because the duty-bearer is the VLOP provider. In both cases the intuition reasoning from subject matter, "this is about AI content and we publish AI-assisted content," pointed the wrong way, and the intuition reasoning from role pointed the right way first. Worth encoding as a reading rule for the next instrument: find the duty-bearer noun before reading the duty. The Political Advertising Regulation is the one miss here that is not role-based; it simply contains no such duty.

For positioning, this tightens the claim the parent brief identified rather than extending it. The sellable observation was "your AI-assisted content is fine, and here is the two-line thing you need on your site to keep it fine." Three instruments later, that claim survives a deliberate attempt to break it, which is the kind of thing worth saying in a market where most coverage is still aimed at the machine-readable marking duty that belongs to model providers rather than to publishers. The re-frame holds: the regulation everybody read as a technology requirement is an editorial-accountability requirement, and the two instruments you would reach for next to find a technology requirement do not contain one. The thing to drop from the pitch is any suggestion that the DSA adds exposure for ordinary publishers. It does not, and saying otherwise would be the fear-selling the vault's own editorial standard rules out.

Why this is in the vault

This closes the third open follow-up from [[2026-09-03-ai-act-article-50-transparency-guidelines]] and confirms, against three primary texts plus the five guideline interplay sections, that the single live action on a shipping RDCO surface, Sanity Check's publicly findable editorial-responsibility statement, survives a deliberate attempt to find a competing labelling duty with no editorial exemption. It records first vault coverage of the DSA, the Political Advertising Regulation and the EMFA, supplies the ¶140 clause that makes the Art. 50(4) carve-out available to a publisher who is not a media service provider, and gives the [[revival-strategy]] sponsorship plan a specific regulatory question to put to a lawyer before the first paid issue.

Open follow-ups

Related

Sources

Primary, fetched and read

All EU texts retrieved via the Cellar content-negotiation route documented in the parent brief: curl -H 'Accept: application/xhtml+xml' -H 'Accept-Language: eng' http://publications.europa.eu/resource/celex/<CELEX>, then stripped to plain text locally and read by article. No WebFetch summarisation was used on any legal text.

Secondary

Research notes trail