Retour aux Insights

AI Act Article 50 and CRM: what teams must bring into compliance now

Pierre FrinSeptembre 20266 min de lecture
Bonjour, je suis une IA. Article 50, AI Act Applicable depuis le 2 août 2026 Adobe Campaign : 0 fonctionnalité native Salesforce : 0 fonctionnalité native AI ACT . ARTICLE 50 . CRM . CHATBOT . CONFORMITÉ 2026
Article 50 of the AI Act has been applicable since 2 August 2026. The machine-readable marking obligations for AI-generated content from existing systems come into force on 2 December 2026.

The AI Act doesn't only concern companies that develop AI models. It also concerns those that deploy them, and in the CRM world, that's almost everyone. As soon as a chatbot responds to a customer, a virtual assistant qualifies a lead, or a conversational agent interacts with your contacts, Article 50 applies.

Most CRM teams haven't yet mapped their exposure. And just like the email tracking pixel in July, neither Adobe Campaign nor Salesforce do anything natively to help you comply.

This article describes a piece of legislation from a reading of its text, it is not legal advice. Whether you qualify as a provider or a deployer, and which obligations follow, depends on your exact situation: have it confirmed by a lawyer.

What Article 50 says : in plain terms

Article 50 of the EU AI Act (Regulation (EU) 2024/1689) imposes transparency obligations on providers and deployers of AI systems that interact with users in the EU.

Obligation 1, Chatbots

  • Any AI system in direct interaction with a human must identify itself as AI
  • Unless it is obvious from the context
  • Applicable since 2 August 2026

Obligation 2, AI-generated content

  • Machine-readable marking required for synthetic content
  • Deepfakes and public-interest content first
  • 2 December 2026 for existing systems

A commonly misunderstood point: the AI Act does not require you to label every marketing email written with AI assistance. A commercial email is neither a deepfake nor content published "to inform the public on matters of public interest". What is mandatory is the declaration of AI systems in direct interaction with your contacts.

What's concerned in CRM : concretely

The question to ask for each tool in your stack: does a customer or prospect interact directly with this system without knowing it's an AI?

Concerned, declaration required

  • Customer service chatbots (Einstein Bot, etc.)
  • Virtual assistants on preference centres
  • Lead qualification bots
  • Conversational agents on e-commerce sites
  • AI auto-responders on email or chat

Not concerned by direct declaration

  • AI-assisted marketing emails
  • Internal scoring and segmentation
  • Non-interactive content personalisation
  • Displayed product recommendations

What Adobe Campaign does natively: nothing

Adobe Campaign Classic and Campaign v8 offer no native functionality for managing Article 50 compliance. No AI declaration setting, no automatic disclosure template, no audit of AI interactions in workflows. If you use third-party AI modules connected to Campaign, responsibility for the declaration remains entirely yours.

What Salesforce does natively: almost nothing

Salesforce has integrated Einstein across virtually all its products, engagement prediction, content generation, lead scoring, Einstein Bots. None of these modules automatically handle the Article 50 declaration for you. The compliance of bots and agents deployed to your customers remains your responsibility.

Be careful about who carries the obligation, the text does not say what it is often made to say. Paragraph 1 of Article 50, the one about a chatbot identifying itself, falls on the provider of the system, not on whoever deploys it. Deployers are covered by paragraphs 3 and 4: emotion recognition, biometric categorisation, deepfakes, and text published to inform the public on matters of public interest.

In practice the line moves quickly: if you assemble a conversational agent, put it on the market under your own name and substantially modify its behaviour, you may become a provider within the meaning of the regulation. Establishing which one you are comes before any compliance work.

What non-compliance costs

Article 99 sets the penalties, and its paragraph 4 names, at point g, "the transparency obligations for providers and deployers pursuant to Article 50". The ceiling is EUR 15 000 000 or 3 % of total worldwide annual turnover for the preceding financial year, whichever is higher.

One detail that changes everything for a small company, and that is rarely quoted: paragraph 6 provides that for SMEs and start-ups, each fine is capped at the percentage or the amount, whichever is lower. The logic is reversed. Paragraph 7 adds that the amount takes into account the seriousness of the breach, cooperation with the authorities and the technical measures already in place.

For the record, the EUR 35 000 000 or 7 % figure often quoted concerns the prohibited practices of Article 5, not the transparency obligations.

What teams need to do now

01
Map all AI systems in direct interaction with your contacts
List every tool in your stack that responds to or interacts directly with customers or prospects. This is step zero, without this mapping, you don't know what you're exposing.
02
Add a declaration notice to every conversational system
A clear notice from the first exchange: "You are interacting with a virtual assistant." On Einstein Bot, this is configurable in the conversation trigger. On third-party chatbots, it must be implemented in the first system message.
03
Prepare AI-generated content marking for December
Generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to implement machine-readable marking. The technical solution needs to be defined now, not in November.
04
Document decisions and trade-offs
Document which systems have been addressed, how, and when. In the event of an audit, this traceability matters as much as the compliance itself.

What this changes in CRM practice

The good news: unlike pixel tracking, Article 50 compliance on chatbots is technically straightforward. A disclosure in the first message of a bot is not six weeks of development. It's a configuration.

The real difficulty is organisational. Who is responsible for this mapping in your organisation? The marketing team that uses the bots? The IT department that hosts them? The DPO? As I described in my article on the organisational crisis of pixel tracking, regulation doesn't create problems, it makes existing ones visible.

The question to ask in your team meeting this week: "Which tools in our CRM interact directly with our contacts without identifying themselves as AI?" If nobody can answer within five minutes, you have your priority.

Official sources
Pierre Frin
Consultant CRM · Adobe Campaign · Salesforce · Imagino · Grokium

Un audit de conformité AI Act sur vos outils CRM ?

Cartographie des systèmes IA en interaction avec vos contacts, évaluation de l’exposition, plan de mise en conformité. Cadrage gratuit, sans engagement.

Contact →
Going further
All technical articles Free Adobe Campaign tools Contact me