All In
FR EN
Back to blog
LinkedIn pour Entreprises 17 min read

EU AI Act and LinkedIn: what US companies must do now that it's in force

| By Patrick de Carvalho

Contents

I have run LinkedIn marketing and content operations for companies on both sides of the Atlantic since 2004, and I have never once seen a compliance deadline arrive with less noise than this one. Article 50 of the EU AI Act, the provision that governs disclosure of AI-generated text, images, and chatbot interactions, became applicable on August 2, 2026. The same day, a separate American law, the California AI Transparency Act, went into effect. Two regimes, one week, and almost no coverage aimed at the people actually posting on LinkedIn. This piece follows up on our earlier prep guide to the EU AI Act, now that the rules are actually in force: what changed and what to do about it, without buying anything.

In short: Article 50 of the EU AI Act has required transparency for AI-generated text, chatbots, and deepfakes since August 2, 2026, and it reaches US companies whenever their LinkedIn content or automation is used by people in the European Union, regardless of where the company is headquartered. The California AI Transparency Act took effect the same day but explicitly excludes text from its labeling requirements. The UK has no equivalent law. The single highest-leverage action for a US marketing team is documenting a human editorial review process, because that is what exempts most AI-assisted text from disclosure under EU law, though it does not exempt AI-generated images or video.

What actually changed on August 2, 2026

The EU AI Act, formally Regulation 2024/1689, rolled out in phases starting February 2025. Article 50, the transparency chapter, is the one that touches marketing directly, and it became enforceable on August 2, 2026, the date the European Commission itself lists as the point the regulation "became applicable, with some exceptions." The Commission published compliance guidelines the same day.

The exceptions matter more than the headline. A separate 2026 amendment package, generally called the Digital Omnibus, pushed the deadline for the Act's high-risk systems chapter, the rules covering things like biometric identification, credit scoring, and hiring algorithms, out to December 2027 for the most sensitive categories and August 2028 for AI embedded in regulated products such as lifts and toys. None of that touches a LinkedIn content calendar. What applies to you today is transparency, not the high-risk regime, and conflating the two is the single most common error I have read in the flood of "AI Act compliance" advice published this summer.

Reach decides this, not your headquarters

Here is the assumption I want to kill first, because it is the one that gets US companies into trouble or, more often, wastes their time on rules that were never theirs to worry about.

Article 2 of the regulation sets its territorial scope, and it does not stop at the EU's borders. It extends to providers and deployers located in a third country whenever the output produced by their AI system is used within the Union. A "deployer," in the regulation's own vocabulary, is simply whoever uses an AI system in a professional capacity, as opposed to a "provider," the company that built it. If your marketing team drafts LinkedIn posts with ChatGPT or Claude and those posts reach followers, connections, or ad audiences in France, Germany, Poland, or any other member state, you are a deployer whose output is used in the Union. Your Delaware incorporation does not change that.

The corollary matters just as much: if your LinkedIn presence genuinely never targets, follows, or is followed by anyone based in the EU, article 50 has nothing to say to you. Check your Sales Navigator filters, your Thought Leader Ads targeting, and your follower geography before you spend a single hour on this. Most of what gets sold to American SMEs as "EU AI Act compliance" right now skips that first question entirely.

Three jurisdictions, three different bets

Within one week in early August 2026, three legal systems staked out three different positions on the same question, and they do not agree.

California (CAITA) United Kingdom European Union (article 50)
Covers AI-generated text? No, explicitly excluded from labeling duties No statute at all Yes, for matters of public interest
Who is obligated Providers of GenAI systems with over 1,000,000 monthly users Nobody, no law Deployers publishing the content
Penalty $5,000 per violation, per day Not applicable Up to €15 million or 3% of worldwide turnover
Effective date August 2, 2026 Not applicable August 2, 2026

California's law, Senate Bill 942 as delayed and amended by Assembly Bill 853, defines a "GenAI system" broadly enough to include text generators. But its actual disclosure duties, the machine-readable "latent disclosure" and the optional visible "manifest disclosure," apply only to image, video, and audio output. Text is named in the definition and then carved back out of the obligation two sections later. And the duty falls on "covered providers," meaning companies that build a public generative AI system used by more than a million people, not on a marketing team that happens to use one. Unless your company operates its own AI writing tool at that scale, CAITA has almost nothing to say about your LinkedIn content.

The UK has legislated nothing on this. The Department for Science, Innovation and Technology lists AI content labeling as a priority area, without a timetable, according to a 2026 House of Commons Library research briefing. What the UK consultation did produce, and it is more useful than a statute, is a clean conceptual line between content that is wholly AI-generated and content that is AI-assisted. That is precisely the distinction American commentary keeps collapsing, and precisely the one the EU law actually turns on.

Article 50, translated into what you post, message, and automate

Strip the legal language and article 50 breaks into four separate duties, and they do not all fall on the same party or carry the same exemptions. I have watched three separate law-firm summaries blur this distinction this summer, so it is worth being precise.

Chatbot disclosure (article 50(1)). If a person interacts with an AI system, they must be told, unless it is obvious to a reasonably well-informed person that they are talking to a machine. This obligation sits with the provider of the chatbot tool, but if you deploy a third-party bot on your LinkedIn landing page or in your DM automation, you inherit the practical risk of it staying silent.

Synthetic content marking (article 50(2)). Providers of tools that generate audio, image, video, or text must embed machine-readable markers in their output. This is a duty on OpenAI, Anthropic, Midjourney, and similar companies, not on you, and it does not apply where the AI performs a merely assistive editing function that leaves the input substantially unchanged.

Deepfake disclosure (article 50(3)). If you deploy AI-generated or manipulated image, audio, or video content that depicts real people, places, or events realistically, you must disclose that it is artificial. A narrow exemption exists for content that is evidently artistic, satirical, or fictional.

Text on matters of public interest (article 50(4)). If you, as a deployer, publish AI-generated text intended to inform the public on a matter of public interest, you must disclose that it was generated or manipulated.

Only the fourth duty carries the human-review exemption everyone quotes. The third does not, and that asymmetry is the part of the law I have seen skipped in every general-audience explainer I read this summer.

The exemption that does most of the work, and the one place it doesn't reach

Article 50(4) does not apply where the content has gone through a process of human review or editorial control, and a natural or legal person holds editorial responsibility for its publication. Read that twice, because it is the load-bearing sentence of this entire compliance question. A US marketing lead who drafts posts with AI, reads them, edits them, and publishes them under the company's name has, in most readings of the text, already satisfied the exemption, provided that process is real and provided you can show it happened. That line, between AI-assisted and AI-generated text, is the same one we unpack in detail elsewhere.

That last clause is the one companies skip. "We review our AI drafts" is not evidence. A dated log entry naming who reviewed a specific post, what was changed, and who approved publication is evidence. It costs nothing and takes ten minutes a week.

The exemption does not travel to article 50(3). If your marketing department uses an AI video tool to generate a "spokesperson" clip, or runs executive headshots through an AI enhancement tool that renders a realistic but synthetic likeness, the editorial-review defense that protects your blog posts does not protect that video. Disclosure is required regardless of how carefully a human reviewed the final cut. I have not seen this distinction stated plainly anywhere else this summer, and it is the reason a company that feels compliant on text can still be exposed on its video content.

The open question nobody has answered yet

"Matters of public interest" is the phrase carrying the whole weight of article 50(4), and the regulation does not define it. In European media law, the phrase has historically leaned toward journalism and civic topics, health, elections, public safety, not a company's product update or a sales pitch. Read narrowly, most routine B2B marketing content on LinkedIn, the kind announcing a hire, a feature release, or a client win, may fall outside article 50(4) entirely, exemption or not, simply because it was never the kind of content the article was written to reach.

Read broadly, any post addressing a genuinely public topic, a regulatory change, a labor market trend, a security incident, could qualify the moment it is aimed at informing an audience rather than selling to it. No enforcement guidance, court ruling, or finalized Code of Practice had settled this distinction as of this article's publication. I am not going to pretend otherwise. Treat the editorial-review process as your baseline regardless of which reading turns out to be correct: it costs nothing, and it is the one action that protects you under either interpretation.

The compliance checklist: eight things, none of them a purchase

  1. Map your AI touchpoints. List every tool that drafts, edits, generates, or automates anything that reaches LinkedIn: writing assistants, image generators, video avatars, chatbots, DM sequencers, lead-scoring add-ons.
  2. Confirm you are actually in scope. Check whether your LinkedIn audience includes people located in the EU, through ad targeting, follower geography, or gated content visitors. No EU reach means article 50 does not apply to you, whatever your headquarters says.
  3. Write down your editorial review process for text. Name who reviews each AI-assisted draft. Note what gets changed. Log who signs off, and date every entry. This is the single action that satisfies article 50(4) in most cases.
  4. Treat AI-generated video and images as a separate category. No editorial-review exemption covers them under article 50(3). If a piece of content shows a realistic depiction of a person, place, or event and was generated or altered by AI, label it, regardless of how much human polish went into it.
  5. Check your chatbots and DM automation for a disclosure step. Anyone interacting with an automated system on your LinkedIn funnel should be told so, unless it is genuinely obvious.
  6. Stop worrying about CAITA unless you build AI tools. California's law targets companies that create a public generative AI system with over a million monthly users. A marketing team using ChatGPT is not that company.
  7. Apply the EU standard in the UK too. No UK statute requires it, but the DSIT has flagged labeling as a coming priority, and the editorial-review process you build for Europe costs nothing extra to run everywhere.
  8. Know your penalty bracket before you estimate risk. EU transparency fines run up to €15 million or 3% of worldwide turnover, whichever is higher for large companies, and whichever is lower for SMEs under article 99(6). Most companies reading this fall into the second bracket.

What LinkedIn's own product team already told you

On July 30, 2026, LinkedIn retired its own AI writing feature, "enhance your post," and replaced it with a tool its team described as one that proofreads a member's words without changing their voice, a move that follows LinkedIn's broader crackdown on AI-generated content. That sentence is worth repeating here because it lands squarely on the same logic as article 50(4): the thing that matters is not whether a machine touched the text, it is whether the author kept the pen. LinkedIn's product team reached the same conclusion European lawmakers did, independently, in the same week. That is not a coincidence I expect to be the last one.

What an executive does Monday morning

Pull up your last twenty LinkedIn posts and video clips that reached an EU audience. For the text, ask whether you could produce a dated record of who reviewed each one. For any video or image with a realistic human likeness, ask whether it says anywhere that it was AI-generated. If both answers hold, you are further along than most of the companies currently paying a consultant to tell them the same thing.

Build the review log this week, in whatever tool you already use for content approval. Nothing about this requires new software, a legal opinion, or a delay to your publishing calendar. It requires a habit: name the reviewer, note the change, date it, keep it.

What I don't know

No enforcement action under article 50 has been brought as of this writing, so nobody, including the lawyers charging for opinions on this, actually knows how narrowly or broadly a regulator will read "matters of public interest" in practice. I don't know whether the Commission's Code of Practice on marking AI-generated content, whose drafts circulated through late 2025 and into 2026, has reached a final version that sets a concrete technical standard, or what that standard will require beyond what article 50 already states. I don't know whether other US states will follow California's lead with laws that treat text differently, or the same way. I have found no evidence that any US company has been contacted by an EU authority over LinkedIn content specifically, and I am not implying that one has. If you read a claim that enforcement has already started against a marketing team, ask for the source before you repeat it. If a Code of Practice update or an enforcement case surfaces, I will cover it in the weekly All In newsletter.

FAQ

Does the EU AI Act apply to a US company that has never sold anything in Europe?

It can, if the company's AI-assisted content reaches people located in the EU. Article 2 of the regulation extends to deployers established in a third country whenever the output of their AI system is used within the Union. A US company whose LinkedIn posts, ads, or chatbots reach EU-based followers or prospects falls within that scope even without a European sales office.

Do I have to label every LinkedIn post I wrote with ChatGPT?

Not necessarily. Article 50(4) exempts AI-generated text from disclosure when the content underwent human review and a named person or company holds editorial responsibility for publishing it. A marketing team that drafts with AI, edits the result, and publishes under its own name has generally met that bar, provided the review is documented.

What's the difference between the EU AI Act and the California AI Transparency Act for text content?

The EU law covers AI-generated text published on matters of public interest, with an exemption for human-reviewed content. California's law defines generative AI broadly enough to include text tools, but its actual labeling duties apply only to image, audio, and video output, and only to companies that build a public AI system with over a million monthly users, not to companies that use one.

Is there a UK law requiring AI content labels?

No. As of 2026, the UK has no statutory obligation to label AI-generated content. The Department for Science, Innovation and Technology has listed labeling as a priority area without a published timetable, according to a 2026 House of Commons Library research briefing.

What counts as "matters of public interest" under article 50?

The regulation does not define the phrase, and no enforcement precedent has settled it. It most plausibly draws on its use elsewhere in European media law, where it has leaned toward journalistic and civic content rather than routine commercial marketing, but that reading has not been tested. Building a documented editorial review process protects you regardless of how the phrase is eventually interpreted.

What are the penalties if I get it wrong?

Under article 99, transparency violations can reach €15 million or 3% of worldwide annual turnover, whichever is higher for large companies. Article 99(6) caps the fine at whichever amount is lower for small and medium-sized enterprises, which covers most companies reading a checklist like this one.

Does the human-review exemption cover AI-generated video or images too?

No. The exemption in article 50(4) applies specifically to text published on matters of public interest. Article 50(3), which governs deepfakes, AI-generated or manipulated content that realistically depicts a real person, place, or event, carries no equivalent human-review carve-out. Disclosure is required regardless of how much editorial oversight the final content received.

Sources

  1. European Commission, "AI Act, Regulatory framework for AI," digital-strategy.ec.europa.eu, consulted August 2026, application timeline and Digital Omnibus postponement to December 2027 and August 2028. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
  2. EU Artificial Intelligence Act, "Article 2: Scope," artificialintelligenceact.eu, territorial reach to third-country providers and deployers. https://artificialintelligenceact.eu/article/2/
  3. EU Artificial Intelligence Act, "Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems," artificialintelligenceact.eu. https://artificialintelligenceact.eu/article/50/
  4. EU Artificial Intelligence Act, "Article 99: Penalties," artificialintelligenceact.eu, fine thresholds and SME reduction under article 99(6). https://artificialintelligenceact.eu/article/99/
  5. Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence, eur-lex.europa.eu. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
  6. California State Legislature, Senate Bill 942 (2024), California AI Transparency Act, leginfo.legislature.ca.gov, covered provider definition, exclusion of text from manifest and latent disclosure duties, $5,000 penalty per violation per day. https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202320240SB942
  7. California State Legislature, Assembly Bill 853 (2025), leginfo.legislature.ca.gov, operative date delayed to August 2, 2026, platform obligations from January 1, 2027. https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB853
  8. House of Commons Library, research briefing CBP-10467, 2026, UK labeling policy status and DSIT priority areas. https://commonslibrary.parliament.uk/research-briefings/cbp-10467/
  9. TechCrunch, "LinkedIn adds a button to report AI-generated slop," techcrunch.com, July 30, 2026, LinkedIn's retirement of "enhance your post" and its replacement described as a tool that proofreads without changing the author's voice. https://techcrunch.com/2026/07/30/linkedin-adds-a-button-to-report-ai-generated-slop/

All In: compliance built on a habit, not a tool purchase

The companies that will handle this well are the ones already keeping a record of who wrote what and why, not the ones shopping for a new AI detector.

All In is the B2B media that decodes LinkedIn, expert blog, weekly podcast and newsletter for SME leaders and sales directors who want to turn LinkedIn into measurable growth. An original creation by Patrick de Carvalho, on LinkedIn since 2004. Motto: "I Never Lose."

Discover All In and get what matters on LinkedIn, every week.