Life Sciences Is Merging Two EU AI Act Obligations Into One

Most life sciences AI governance programs are built around one question: does this system meet the risk threshold for FDA clearance or MDR classification. The FDA itself is still working out that threshold for generative tools, with officials signaling more guidance is coming and continuing to weigh its regulatory approach to generative AI in medical devices. That question matters. It is also not the only one on the table, and compliance teams are increasingly collapsing two separate EU AI Act transparency obligations into a single undifferentiated bucket.

Two obligations, not one

The AI Act’s transparency provisions cover distinct scenarios. One obligation concerns synthetic content itself: outputs need to carry a machine-detectable marker so audiences and platforms can identify AI-generated material, a requirement that has drawn scrutiny over how watermarking gets implemented in practice. This is the obligation that touches a pharmaceutical marketing team drafting patient education material or a medical affairs group producing multilingual disease-awareness content. The content itself, not the system generating it, is what triggers the requirement.

A second obligation concerns interaction, and it is a different thing entirely. When a system is designed to converse with a person, the person needs to know they are talking to AI, independent of whether any content produced in that exchange gets marked or shared afterward. Healthcare systems adopting cloud-based conversational voice AI for scheduling, insurance verification, and access workflows are building squarely into this second obligation. The voice AI is not seeking 510(k) clearance and will never touch a device file, but the disclosure logic still applies, and it applies through a different mechanism than watermarking a piece of content.

Treating these as one compliance track invites a specific failure mode: a program that builds watermarking into its content pipeline and assumes it has covered “transparency,” while its call-handling AI never tells a single patient it is a machine.

Where the professional layer sits

Practice-level bodies are starting to fill governance gaps that sit above the device question entirely. The Pharmacists’ Defence Association’s new AI guidance frames AI use in pharmacy as carrying professional, clinical, and legal considerations that exist alongside product-level regulation, not inside it. That instinct is correct, but it also underscores how many distinct obligations now sit outside the MDR and FDA classification frame that most compliance playbooks were built around.

The decision in front of sponsors

The risk is not enforcement exposure alone. It is discovering, mid-deployment, that a content-marking fix does nothing for an interaction-disclosure gap, or vice versa. Sponsors operating in the EU should inventory every generative AI touchpoint that reaches a patient or public audience and classify each one against both obligations separately: does it produce synthetic content that needs marking, does it interact with a person who needs to be told, or both.

The device pathway will get its guidance in time. The transparency layer already has two rules, not one, and conflating them is the gap worth closing now.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core distinction between content-marking and interaction-disclosure obligations is legally sound and the argument is coherent, but the piece asserts this conflation is happening (‘compliance teams
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but the article could benefit from more specific references to the EU AI Act provisions.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately distinguishes the two EU AI Act transparency obligations (content marking vs. interaction disclosure) and aligns with ISO 42001’s emphasis on transparency and risk management,
Technical Accuracy	Llama	cleared. The article accurately distinguishes between two separate EU AI Act transparency obligations and correctly identifies their distinct applications in the life sciences industry.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and addresses a potential conflation of two distinct EU AI Act obligations, providing clear counterarguments to the ‘one compliance track’ approach and avoiding ven
Novelty & Non-Duplication	Grok	cleared. The two-obligation distinction (content marking vs interaction disclosure) is a genuine synthesis not present as such on the wire; individual source items are commodity, but the conflation failure-mod
Validation	DeepSeek	cleared. The central claim that the EU AI Act contains two distinct transparency obligations is factually correct and validated by the Act’s text.

Sources cited: 15. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.