The EU AI Act Is Now Live for Sales Teams: What Applies Since August 2026
Since August 2, 2026 the EU AI Act's transparency rules are enforceable, and every team using AI chatbots, AI-drafted content or AI voice agents in sales and marketing is in scope. What Article 50 actually requires, what got postponed to 2027 and 2028, why human-reviewed outreach sits comfortably, and the five-point check to run this quarter.
Key takeaways
- Since August 2, 2026 the AI Act's Article 50 transparency obligations are enforceable: people must be told when they interact with an AI system, synthetic media must be machine-readably marked, and AI-generated text published to inform the public needs disclosure. This applies to ordinary companies, not just AI vendors.
- The heavy high-risk obligations were postponed via the Digital Omnibus to December 2027 and August 2028, and pre-existing generative systems have until December 2, 2026 for machine-readable marking. The panic headlines and the actual duties differ a lot.
- For outbound the key line: one-to-one sales emails drafted by AI and substantively reviewed by a human are not labeling-covered public-interest content. The human checkpoint that quality already demanded is also the compliance-friendly posture.
- The concrete to-do list is short: inventory your AI touchpoints, add a disclosure line to chatbots and voice agents, document human review of outgoing content, ask your tool vendors about marking compliance, and diarize the December 2026 date.
The EU AI Act stopped being a future topic on August 2, 2026: since that date its transparency obligations, Article 50, are in force and enforceable, and they apply to ordinary companies using AI, not only to the vendors building it. At the same time, most of what the scare-headlines described never arrived, because the heavy high-risk obligations were postponed. Sales and marketing teams now sit in the odd middle: some duties are real today, most are not, and telling them apart is worth twenty minutes. This is that twenty minutes. One caveat up front: this is a practitioner's orientation, not legal advice; the specifics of your setup belong in front of your counsel.
What actually changed on August 2, 2026
Article 50 covers four situations: people interacting directly with an AI system, AI-generated synthetic content (audio, image, video, text), emotion recognition and biometric categorization, and deepfakes plus AI-generated text published to inform the public. The duties split between providers, the companies building the AI system, and deployers, the companies using it, which is where most sales teams sit.
Equally important is what did not arrive. The high-risk obligations, the ones with conformity assessments and documentation regimes, were pushed back via the Digital Omnibus to December 2027 and August 2028. And generative systems already on the market before August 2026 have a grace period until December 2, 2026 for the machine-readable marking duty. So the enforceable core today is transparency, not certification.
What it means for a sales and marketing motion, concretely
Chatbots and voice agents: disclose, always. If your website chat, your inbound qualification bot or an AI voice agent talks to prospects, they must know they are talking to AI, clearly, at the latest at first interaction. There is an exemption when AI involvement is obvious to a reasonably informed person, but building on that exemption is exactly the kind of cleverness that ages badly. A one-line disclosure costs nothing; add it.
AI-drafted outreach: the human gate carries you. One-to-one B2B emails and LinkedIn messages are not "text published to inform the public on matters of public interest", and the labeling duty for public-interest text explicitly does not apply where content underwent substantive human review with editorial accountability. In other words: the checkpoint model we have argued for on quality grounds in AI agents in sales, a human reviewing everything that leaves under your name, is also the posture that keeps outreach comfortably outside the labeling debate. Fully autonomous sending, on top of every quality and deliverability argument against it, now also carries a regulatory question mark it does not need to carry.
Synthetic media in marketing: know your vendor's duty and your own. The machine-readable marking of AI-generated images, video and audio is primarily the provider's job, your generation tool must embed it, and the technical standards are being finalized through a Code of Practice. Your job as a deployer is narrower: do not strip markings, disclose deepfake-style content, and if you publish AI-generated text to inform the public, label it or run real editorial review. Practical consequence: "does your output comply with Article 50 marking" is now a standard question for every content tool in your stack, right next to the MCP question from our tool evaluation habit.
What not to panic about. Ordinary CRM analytics, lead scoring on firmographic data, research agents and drafting assistants are not in the enforceable scope today. The channel rules for first contact in DACH, competition and data protection law, remain the stricter and older constraint on outbound, unchanged by the AI Act; the per-country details are in our guide on whether cold outreach is allowed.
The five-point check for this quarter
First, inventory every point where AI touches a prospect: chat widgets, voice agents, generated content, drafted outreach. Second, add the disclosure line to anything conversational. Third, write down your review process for outgoing content, who reviews, what accountability means, because the exemption for human-reviewed text assumes real editorial control, not a rubber stamp. Fourth, email your tool vendors the marking question and keep the answers. Fifth, put December 2, 2026 in the calendar: that is when the marking grace period for pre-existing systems ends, and a good moment to re-check the inventory.
The honest summary
For a disciplined outbound team, the AI Act's live obligations confirm the setup you should already have: humans reviewing what goes out, disclosure where a machine talks to a human, and vendors who can answer compliance questions. Teams running fully autonomous AI outreach have a new reason to reconsider, on top of the old ones. And the competitive angle is quietly useful: being able to tell a compliance-minded DACH prospect exactly how your AI-assisted process meets the transparency rules is trust-building content for a market that takes exactly these questions seriously.
Frequently asked questions
What does the EU AI Act require from sales teams since August 2026?
The enforceable core is Article 50 transparency: people must be informed when they interact with an AI system such as a chatbot or voice agent, AI-generated synthetic media must carry machine-readable marking (primarily the tool provider's duty), and AI-generated text published to inform the public needs disclosure unless it underwent substantive human editorial review. Heavy high-risk obligations were postponed to December 2027 and August 2028.
Do AI-drafted cold emails need to be labeled as AI-generated?
One-to-one B2B outreach is not public-interest publishing, and the text-labeling duty explicitly gives way where content receives substantive human review with editorial accountability. An AI-drafted, human-reviewed and human-sent email therefore sits outside the labeling debate. Fully autonomous AI sending is the setup that raises questions, and it was already the wrong setup for quality and deliverability reasons.
Must website chatbots disclose that they are AI?
Yes. Since August 2, 2026, people interacting with an AI system must be clearly informed at the latest at first interaction. There is an exemption where AI involvement is obvious to a reasonably informed person, but relying on it is poor practice; a one-line disclosure in the chat window resolves the duty cleanly and costs nothing.
Were the EU AI Act high-risk rules postponed?
Yes. Via the Digital Omnibus, the heavy high-risk obligations were deferred to December 2027 and August 2028, and generative AI systems already on the market before August 2026 have until December 2, 2026 to comply with machine-readable marking. What is enforceable now for ordinary companies is the transparency layer, not certification regimes. Details of any specific setup belong with legal counsel.