If your team sells with a website chatbot, an AI SDR, an AI voice agent or AI-generated video, the EU AI Act now applies to how you sell. Since 2 August 2026, Article 50 of the Act requires AI that talks to people to say it is AI, and deepfakes such as cloned voices or avatar videos to be labelled.
In Salesforce's seventh State of Sales report, a survey of more than 4,000 sales professionals in 22 countries, 54% of sellers said they have used AI agents, and nearly 9 in 10 plan to by 2027.
Below: the timeline, what Article 50 and the AI literacy duty in Article 4 require in plain language, the fines and a checklist. This is not legal advice, so confirm your case with counsel.
When does the EU AI Act apply? The timeline for sales teams
Most of the AI Act has applied since 2 August 2026. The Act, Regulation (EU) 2024/1689, comes in stages, as the European Commission's AI Act page sets out:
- 1 August 2024: the AI Act entered into force.
- 2 February 2025: bans on prohibited AI practices and the AI literacy duty.
- 2 August 2025: rules for general-purpose AI models, which many AI tools are built on, and governance.
- 2 August 2026: general application, including Article 50, with enforcement by the AI Office and national authorities.
- 2 December 2027 and 2 August 2028: high-risk rules, first for areas such as employment and education, then for AI in regulated products such as lifts or toys.
The last line is new. The AI Omnibus Regulation (EU) 2026/1744, proposed by the Commission on 19 November 2025 and in force since 27 July 2026, pushed the high-risk rules back. It did not move Article 50; it only gave providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to add machine-readable marking.
One exception to watch: AI that monitors and evaluates the performance of your own reps can fall in the employment area of the high-risk list in Annex III.
What AI Act Article 50 requires, in plain language
Article 50 splits the work between the provider, the company that builds an AI system and offers it under its own name, and the deployer, the company that uses it, which is usually you. The text of Article 50 sets four duties:
- AI that talks to people must say so. The provider must design the system so people know they are dealing with AI, unless that is obvious to a reasonably well-informed, observant and circumspect person.
- AI-generated content must be marked. Providers of generative AI must mark audio, image, video and text output in a machine-readable format.
- Emotion recognition must be disclosed. Deployers must tell people when an emotion recognition or biometric categorisation system is used on them.
- Deepfakes must be labelled. Deployers must disclose them, and also AI text published to inform the public on matters of public interest, unless a person reviewed it and holds editorial responsibility.
The information must be clear and reach people at the latest at the first interaction or exposure. The Commission's Article 50 Q&A says the duty covers people 'whether consumers, professionals or other users', so B2B buyers count, and that the 'obvious' exception is to be read restrictively.
For most tools you buy, the vendor is the provider. But the Commission's guidelines on Article 50 of 20 July 2026, which are non-binding, treat a company that builds a chatbot in-house and runs it under its own name as a provider itself. A company also stays the deployer when contractors run a system on its behalf and under its control.
Also check your call-analysis tools: using AI to read the emotions of your own staff at work has been banned since 2 February 2025, except for medical or safety reasons.
AI chatbots, AI SDRs and AI voice calls: how to disclose
The test in the guidelines: if the AI itself talks or writes to a person in a real two-way exchange, by chat, phone or email, disclose it at the start. The exchange does not have to be started by a human, so an AI's first outreach message counts.
- Website chat. Disclose in the first message and keep a visible AI label. A line in your terms, or a vague name such as 'assistant', is not enough.
- AI voice calls. Say in the first sentence that this is an AI assistant and which company it calls for, and remind people in longer calls. A human-sounding voice makes the 'obvious' exception harder to rely on.
- AI SDRs that email on their own. Agents must disclose that they are AI and on whose behalf they act, for example with an AI label at the top of the email.
- AI drafts that a rep reviews and sends. This counts as indirect use, so the email needs no Article 50 label. The review has to be real: the mere possibility of human review does not remove the duty.
- Direct questions. If a prospect asks whether they are talking to a bot, the system has to say so.
Disclosure costs one sentence; a buyer who later finds out the 'rep' was a bot loses trust in you. For which work to give AI at all, see AI SDR or outsourced sales team and our note on Salesforce's Hunter outbound agent.
Voice clones and avatar videos: labelling deepfakes in outreach
The Act defines a deepfake as AI-generated or manipulated image, audio or video that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. The guidelines list a realistic synthetic avatar of a company CEO as an example, so an avatar video of your CEO or a cloned rep voice in outreach very likely needs a label.
Your company, as deployer, must disclose it at the latest at first exposure, with a visible or audible label. According to the Commission's Q&A, the machine-readable watermark your video tool adds is not enough. If the person shown or heard is real, data protection law and personality rights apply too, so get their written agreement first. Content made before 2 August 2026 needs no retroactive label.
AI literacy under Article 4: what your sales team needs
Article 4 has applied since 2 February 2025, and the AI Omnibus rewrote it from 27 July 2026. Instead of ensuring 'a sufficient level of AI literacy', providers and deployers must now take measures to support the AI literacy of their staff and of other persons who use AI systems on their behalf, without having to guarantee any specific level. 'On their behalf' can include freelancers and agencies. A proportionate answer for a sales team:
- A short session on what your AI tools do, where they get things wrong and what has to be disclosed.
- One page of rules: approved tools, what data may go into them and when a person checks output before it goes out.
- A record of who was briefed and when, including outsourced partners.
This works best inside regular coaching, not as a one-off session; our IT sales training page explains how we coach sales teams.
EU AI Act fines: what Article 99 says
Breaching the Article 50 duties can cost up to 15 million euros or 3% of total worldwide annual turnover, whichever is higher, under Article 99. For SMEs, start-ups and, since the AI Omnibus, small mid-cap companies, the lower of the two applies. Prohibited practices carry up to 35 million euros or 7%.
These are ceilings. Member States set the penalty rules, national market surveillance authorities do most of the enforcement, and fines take account of company size, duration, cooperation and intent.
EU AI Act checklist for sales teams
Work through this with whoever owns your sales tools.
| Tool or practice | What applies | What to do |
|---|---|---|
| Website chatbot | Art. 50(1): disclose from the first message | Disclosure in the first line, a visible AI label, an honest answer when asked |
| AI voice agent or AI calling | Art. 50(1): spoken disclosure at the start | Say it is AI and for which company; remind people in long calls |
| AI SDR that emails on its own | Art. 50(1): disclose AI and whom it acts for | An AI label in the email, or a person who reviews and sends each reply |
| Voice clones and avatar videos | Art. 50(4): deepfake label at first exposure | A visible or audible label; written agreement of the person shown |
| Staff and agencies using AI | Art. 4: measures to support AI literacy | Short training, one page of rules, a record of who was briefed |
| Buying AI sales tools | Provider duties under Art. 50(1) and 50(2) | Ask vendors in writing how they disclose and mark output; keep the answers |
Use the last row when you compare vendors, for example from our guide to AI sales tools for startups, and ask whether the disclosure can be switched off. Keep a person in the loop wherever a conversation creates a commitment, such as pricing or contract terms, and give one person ownership of this list.
The email and phone rules that still apply
The AI Act adds to the rules you already follow; the AI Omnibus text states that it does not affect the GDPR. These still apply, whether a person or a model wrote the message:
- GDPR. Recital 47 says processing personal data for direct marketing may be regarded as a legitimate interest, which still needs a balancing test and an easy way to object.
- Czech Republic. Commercial email needs the recipient's consent in advance, and the rule protects companies too, with an exception for existing customers and similar products, as the Czech data protection office explains. Germany, Poland, Italy and six other jurisdictions set their own email rules, and our guide to B2B cold email law across Europe compares them.
- France. Since 11 August 2026, Article 13 of Law No. 2025-594 bans telephone canvassing of a consumer who has not consented in advance, except about an ongoing contract. The Bloctel list ceased to exist that day and contracts from non-compliant calls are not valid, as Service Public explains. Fines reach 375,000 euros for a company.
The French rule protects consumers, which French consumer law defines as natural persons acting outside their trade or profession, so calls to companies about their business fall outside it. An AI voice agent calling French consumers needs both: prior consent and an AI disclosure.
The AI Act does not change deliverability rules either. If replies are dropping, see why cold emails go to spam, and for building compliant lists, where to find and qualify IT sales leads.
Frequently asked questions
Does the EU AI Act apply to B2B sales?
Yes. Article 50 protects natural persons, and the Commission's Q&A names consumers, professionals and other users. A buyer at a company is a natural person, so an AI chatbot or voice agent that talks to them must say it is AI. Providers outside the EU are covered too when their system's output is used in the EU.
Do AI-written sales emails need an AI label?
Not when a rep drafts with AI, reviews the text properly and sends it as the person communicating; the Commission's guidelines treat that as indirect use. An AI agent that sends and answers email on its own is different: it should disclose that it is AI and whom it works for, for example with a label at the top of the email.
Did the AI Omnibus delay Article 50?
No. The AI Omnibus, in force since 27 July 2026, moved the high-risk rules to 2 December 2027 and 2 August 2028 and softened the AI literacy duty. Article 50 applies from 2 August 2026, with one grace period: generative AI systems already on the market before that date have until 2 December 2026 for machine-readable marking.
Use AI in sales and keep the buyer's trust
The AI Act does not stop you using AI in sales. It asks you to be open when a machine talks to a person, and to train the people who use the tools.
IT SalesaaS runs outsourced B2B sales for IT and SaaS companies in English, Czech and Turkish. If you want a sales process that uses AI where it helps and keeps people where they matter, see sales strategy development and lead generation, and what each plan includes on our pricing page.
This reflects the rules as of 27 September 2026 and is not legal advice; check your case with counsel.
Sources and further reading: European Commission, AI Act overview; European Commission, Article 50 Q&A; European Commission, Article 50 guidelines; EUR-Lex, the AI Act; EUR-Lex, the AI Omnibus Regulation; AI Act, Article 50 text; AI Act, Article 4 text; AI Act, Article 99 text; AI Act, Annex III high-risk list; Salesforce, seventh State of Sales report; GDPR Recital 47; ÚOOÚ, Act No. 480/2004; Légifrance, Law No. 2025-594, Article 13; Légifrance, Consumer Code Article L242-16; Légifrance, consumer definition; Service Public, telephone canvassing rules. To follow new AI Act guidance and enforcement as it appears, a source of AI news from primary sources helps you check each claim against the original text.