ClearLabel → Vendors → Salesforce Embedded Service / Agentforce
Salesforce Embedded Service / Agentforce and EU AI Act Article 50
Only if the AI mode is switched on — and you are the one who has to check.
ClearLabel → Vendors → Salesforce Embedded Service / Agentforce
Only if the AI mode is switched on — and you are the one who has to check.
Salesforce Embedded Service / Agentforce ships both human-routed and AI-agent modes. An outside scan cannot see which one your workspace runs. If the AI answers first, Article 50(1) applies to you as the deployer, not to the vendor. If a human always answers, it does not. Either way, write down which it is and when you checked — that record is your evidence.
AI product to check: Agentforce. Agentforce agents act autonomously. Disclose in the deployment's agent alias.
Embedded Service Deployment → Chat Settings → agent alias; Agentforce → persona.
Article 50(1) requires the information to be given clearly and distinguishably, at the latest at the time of the first interaction. A line buried in your privacy policy does not satisfy that. It has to be where the conversation starts.
en | You're chatting with an AI assistant. It can make mistakes — ask for a human at any time. |
de | Sie chatten mit einem KI-Assistenten. Er kann Fehler machen — fragen Sie jederzeit nach einem Menschen. |
fr | Vous discutez avec un assistant IA. Il peut se tromper — demandez un conseiller humain à tout moment. |
es | Estás hablando con un asistente de IA. Puede cometer errores — pide hablar con una persona cuando quieras. |
it | Stai parlando con un assistente IA. Può commettere errori — puoi chiedere un operatore umano in qualsiasi momento. |
nl | Je chat met een AI-assistent. Deze kan fouten maken — vraag altijd om een medewerker. |
pl | Rozmawiasz z asystentem AI (sztuczna inteligencja). Może on popełniać błędy — w każdej chwili możesz poprosić o rozmowę z człowiekiem. |
pt | Está a falar com um assistente de IA. Ele pode cometer erros — peça para falar com uma pessoa a qualquer momento. |
sv | Du chattar med en AI-assistent. Den kan göra misstag — be när som helst om att få prata med en människa. |
da | Du chatter med en AI-assistent. Den kan begå fejl — bed når som helst om at tale med et menneske. |
fi | Keskustelet tekoälyavustajan kanssa. Se voi tehdä virheitä — voit milloin tahansa pyytää keskustelua ihmisen kanssa. |
lt | Jūs kalbatės su dirbtinio intelekto asistentu. Jis gali klysti — bet kada galite paprašyti, kad atsakytų žmogus. |
Serve the language your customer is browsing in. Article 50 also requires the disclosure to meet accessibility requirements, so do not deliver it as an image.
2 August 2026 — in force. Art. 50(1) AI-interaction disclosure. Art. 50(4) deepfake and public-interest text disclosure.
2 December 2026. Art. 50(2) machine-readable marking of synthetic audio, image, video and text, for systems already on the market before 2 Aug 2026.
2 December 2027 — deferred. Most standalone high-risk (Annex III) duties, pushed back 16 months by the Digital Omnibus approved on 29 June 2026. Article 50 was not deferred.
Penalties for transparency breaches sit under Art. 99(4): up to €15m or 3% of worldwide annual turnover, whichever is higher, with the actual amounts set by each member state.
The scan reads your public HTML, fingerprints Salesforce Embedded Service / Agentforce and 35 other vendors, and looks for disclosure wording in twelve EU languages. Free, no signup, runs in your browser.
Scan your site freeFin AI Agent
MagicReply / Crisp AI
Lyro AI
Breeze / AI Chatbot
Zendesk AI agents
LiveChat AI / ChatBot.com
Only if the AI mode is switched on — and you are the one who has to check. Salesforce Embedded Service / Agentforce ships both human-routed and AI-agent modes. An outside scan cannot see which one your workspace runs. If the AI answers first, Article 50(1) applies to you as the deployer, not to the vendor. If a human always answers, it does not. Either way, write down which it is and when you checked — that record is your evidence.
You are. Article 50(1) places the duty on the provider of the AI system and, for deepfakes and public-interest text under 50(4), on the deployer. In practice, if you enabled the AI mode in Salesforce Embedded Service / Agentforce and pointed it at your customers, the exposure is yours.
Article 50(1) has applied since 2 August 2026. The machine-readable marking duty in Article 50(2) has a transition to 2 December 2026 for systems already on the market. The Digital Omnibus deferred most high-risk obligations to 2 December 2027 but did not defer Article 50.