EU AI Act transparency and GPAI claims: what should buyers ask?
Last reviewed May 30, 2026
EU AI Act transparency claims often appear when a product includes chatbots, AI agents, synthetic content, or user-facing AI interactions. This guide maps those claims, and GPAI model systemic-risk claims, to source-backed questions a buyer can ask before relying on the wording.
Evidence buyers verify
- The exact EU AI Act transparency claim and the feature it applies to.
- The source article, section, or official guidance behind the wording.
- The product role: provider, deployer, distributor, customer-configured workflow, or mixed role.
Opens the checker for this claim type. Paste your vendor's exact wording there. Evidence questions only — not a blacklist or fraud detector. Not sure what a result looks like? See a sample receipt.
Sources this guide draws from
- · 13 June 2024
Official legal text source for Article 50 transparency obligations and role boundaries.
- · Accessed May 24, 2026
European Commission service desk source summarizing Article 50 transparency obligations and clear-accessible notice requirements.
- · Accessed May 24, 2026
Public company source for user-facing AI agent and chatbot capability wording; used as claim wording evidence, not EU AI Act compliance evidence.
- · Published December 11, 2025
Public company source for AI agent front-line support and human handoff wording; not independent compliance validation.
- · 13 June 2024
Source for GPAI model systemic-risk classification, the 10^25 FLOPs training-compute threshold, and additional obligations for providers above that threshold: model card, transparency report, copyright summary, adversarial testing records, and serious-incident reporting.
Public claims with documented evidence gaps
"Fin AI Agent is more than a chatbot"
Vague AI-powered- Source and date
- Intercom Fin AI Agent FAQs · Accessed May 24, 2026
- Evidence signal
- User-facing AI interaction wording that needs a disclosure point and role boundary.
- Evidence gap
- A buyer needs the user notice, first-interaction disclosure, role of the AI agent, handoff rule, and affected-user context.
- Buyer question
- For the more than a chatbot claim, when is a user told they are interacting with an AI system and what role can the agent play?
"instantly generates answers to resolve customer issues"
Automation / Replacement- Source and date
- Intercom Fin AI Agent FAQs · Accessed May 24, 2026
- Evidence signal
- Generated-answer wording tied to customer-facing support without visible notice, review, or exception boundary.
- Evidence gap
- A buyer needs generated-content handling, source visibility, confidence threshold, escalation logic, and notice shown to users.
- Buyer question
- For the instantly generates answers claim, what disclosure, source reference, and handoff appears before the user relies on the answer?
"Fin will provide instant responses and a robust handover experience"
Automation / Replacement- Source and date
- Intercom Fin AI Agent answer-first workflow · Published December 11, 2025
- Evidence signal
- Front-line AI support claim that depends on workflow setup, handoff design, and user-facing notice.
- Evidence gap
- A buyer needs the handoff conditions, routing rules, transcript context, user notice, and human review boundary for unresolved requests.
- Buyer question
- For the instant responses and handover claim, what conditions trigger handoff and what information is shown to the user and teammate?
Match each claim pattern to the evidence buyers need
| Claim pattern | Evidence needed | Buyer question |
|---|---|---|
| EU AI Act-ready chatbot or AI agent transparency | Article 50 source, provider/deployer role, first-interaction notice, user context, and exception handling. | Where does the product inform users that they are interacting with an AI system? |
| AI-generated content label, synthetic content watermarking, or AI-manipulated content disclosure | Content type, generation path, machine-readable marking, user-facing label, watermark persistence, and editorial review boundary. | How is generated or synthetic content marked, and can the label or watermark survive export, sharing, or downstream publishing? |
| AI support agent, AI assistant, or chatbot interaction | Notice timing, role description, source visibility, escalation path, and user ability to reach a human. | At the first interaction, what does the user see about the AI nature and limits of the system? |
| Emotion recognition, biometric categorisation, or deepfake disclosure | Whether the feature exists, exposed-person notice, processing context, content type, and disclosure placement. | Does the product include any feature that triggers a separate disclosure path beyond standard chatbot notice? |
| GPAI model compliance or EU AI Act systemic-risk threshold claim | Model classification status (above or below the 10^25 FLOPs training-compute threshold), model card, transparency report, copyright summary, adversarial testing records, and confirmation of serious-incident reporting channel. | Is the underlying model classified as a GPAI model with systemic risk under Article 51, and if so, what model card, transparency report, and adversarial testing records does the provider publish? |
Evidence to request
- The exact EU AI Act transparency claim and the feature it applies to.
- The source article, section, or official guidance behind the wording.
- The product role: provider, deployer, distributor, customer-configured workflow, or mixed role.
- Screenshots or copy showing user notice at first interaction or first exposure.
- Generated-content labels, synthetic-content watermarking, machine-readable marking, handoff rules, and exceptions for human editorial review.
Questions to put in front of the vendor
- For this EU AI Act transparency claim, which Article 50 obligation does the vendor map it to?
- Is the product interacting directly with natural persons, generating synthetic content, or supporting a deployer workflow?
- Where is the AI disclosure displayed, and is it clear before the user relies on the interaction or content?
- If the claim mentions generated-content labels, synthetic content, or watermarking, what output types and downstream exports are covered?
- What settings can our team configure without removing the notice, label, or handoff boundary?
- What wording should be narrowed if the vendor only supports one disclosure surface or one customer configuration?
- If the product is built on a GPAI model: is that model above the systemic-risk compute threshold, and what model card, transparency report, and adversarial testing records does the provider publish?
Wording boundaries to compare against
- Supports configurable AI interaction notices for specified chatbot and AI-agent workflows.
- Provides generated-content labels for named output types; downstream publishing controls remain the customer's responsibility.
- Maps selected transparency features to Article 50 source context without stating a full compliance conclusion.
- Documents handoff conditions and user notice for customer-facing AI support interactions.
- States GPAI model classification status and links to published model card and transparency report rather than claiming systemic-risk compliance without documentation.
Have your vendor's exact claim wording ready?
Check an EU AI Act transparency claim How the evidence method works