The FDA Just Said AI Devices Need Lifecycle Proof. Does Your AI-Drafted CER?

If the FDA won't accept "trust the model" as proof for an AI-enabled device, why would you accept it from the AI tool drafting your Clinical Evaluation Report (CER)?

You shouldn't. We think scrutiny of AI inside a device should raise the bar for AI used in the paperwork about that device. That means every claim linked to a source, every draft versioned, and evidence kept continuously, not assembled once at submission. This is our judgement, not an MDR requirement: the regulation does not govern drafting tools, but it does hold you responsible for what the CER says.

What did the FDA actually announce?

On Thursday 1 October 2026 the FDA published its fiscal year 2027 guidance priorities, as MedTech Dive reported the next day. Finalising guidance on marketing submissions and lifecycle management for AI-enabled devices is among the top priorities for the coming 12 months. World Health AI's coverage reports the same.

The document in question is the draft guidance of January 2025. Last year's list treated finalising it as a second-tier goal, so it has moved up.

Two cautions. A priority list is an intention, not a promise, and the guidance stays a non-binding draft until the FDA issues a final version. The FDA is taking comments on its priorities until 30 November 2026.

Two speeds, one principle

The EU is moving more slowly. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force 27 July 2026) moved the AI Act's high-risk date for AI built into regulated products, including medical devices, from 2 August 2027 to 2 August 2028. Law firm Freshfields describes the same shift.

	FDA	EU
Instrument	Draft guidance on AI-enabled device software, lifecycle management and marketing submissions	AI Act high-risk rules for AI in devices (Regulation (EU) 2024/1689)
Status	Draft since January 2025, finalisation prioritised for FY2027	Law, applies from 2 August 2028
Legal weight	Non-binding recommendations	Binding

The speeds differ, but the FDA draft states the principle plainly. It frames its recommendations around managing risk across the total product lifecycle, not only at the moment of submission.

Does the 2028 date give EU teams more runway?

Only for the AI Act layer, and only for the AI inside the device. The MDR applies today. Article 61(11) already requires the clinical evaluation to be updated throughout the device lifecycle with post-market data.

The date also says nothing about the tool that drafts your CER. Nobody has postponed that question, because no rule has asked it yet.

What would lifecycle evidence mean for a CER?

The FDA draft is about the device's AI, not about drafting tools. But the questions it implies carry over well. The table below is our own mapping, not FDA or MDR text.

The lifecycle question	What it means for an AI-drafted CER
Where did this come from?	Each claim links to a real reference and to a recorded screening decision.
Which version produced this?	The draft has a version history, and you record which tool and model version generated it.
What changed, and why?	A change log shows what the AI proposed and what a human edited or rejected.
Does it still hold?	You re-check outputs when the tool, the model or the evidence changes.
Who is accountable?	A named, qualified reviewer signs off conclusions such as benefit-risk.

For a deeper look at the first, third and fifth rows, see our article on responsible, audit-ready AI use in CER drafting. For the regulatory timelines in more detail, see why Washington out-paced Brussels on AI device rules.

Why does "continuously" matter?

A CER is not finished at submission. Under MDR Article 61(11) it is a living document, updated as post-market data arrives. A drafting process that can only be explained on the day you first wrote the report will not be explainable at the next update.

The gap nobody's auditing

Nothing in the MDR asks how a first draft was produced, so nobody is obliged to ask. We cannot say how any given notified body will treat AI-assisted drafting, and views may vary. The accountability does not move, though: the manufacturer owns every sentence.

Here is an illustrative scenario. A team asks a general-purpose chatbot to draft the state-of-the-art section of a CER update. The text reads well, cites seven papers, and passes a quick proofread.

Months later a reviewer asks where one performance figure came from. The chatbot's session is gone, nobody recorded which papers were screened, and one reference cannot be found. Fluent text is not evidence, and general-purpose chatbots can produce references that do not exist.

What is the gut-check for your drafting tool?

Ask your drafting tool for the same thing the FDA expects from a device maker: show your work, continuously, not once at submission. In practice, put five questions to it:

1. Can it link every statement to a source document you have already searched and screened?
2. Does it keep a version history of drafts, edits and approvals?
3. Can you see what the AI proposed and what a person changed?
4. Do you have a written intended use for the tool, and do you re-check it when the model changes?
5. Does a named reviewer sign off each conclusion?

If you answer "no" to any of these, you have found your gap. Fixing it is cheaper before an auditor asks than after.

Key takeaways

• The FDA put finalising its AI-enabled device lifecycle guidance among its top FY2027 priorities, up from second tier last year. It is still a non-binding draft.
• The EU AI Act date for AI in devices moved to 2 August 2028, but the MDR applies now and the CER must already be kept up to date.
• The 2028 date concerns AI inside the device, not the tool that drafts your documents.
• In our view, an AI-drafted CER needs traceable sources, version history, human review and tool re-checks.
• Test your drafting tool the way the FDA tests a device: show your work, continuously.

FAQ

Is the FDA's lifecycle guidance for AI devices final?

No. The FDA published a draft in January 2025 and has made finalising it a top priority for FY2027. Until then it remains non-binding.

Does the EU AI Act regulate the tool I use to write my CER?

The AI Act's high-risk rules for devices cover AI in or as a device. We are not aware of any provision that regulates drafting tools used for regulatory documents, but you stay responsible for the content under the MDR.

Does the FDA guidance apply to my CER?

No. It covers marketing submissions for devices with AI-enabled software functions in the US. We use its lifecycle logic as a useful benchmark for the drafting process, not as a requirement.

When does the AI Act apply to medical devices?

From 2 August 2028 for AI built into products under Annex I, such as devices, following the Digital Omnibus. Check the final legal text for your device type.

Do we need to tell our notified body that AI helped draft the CER?

We know of no MDR provision requiring it. Ask your notified body early and check your quality management system.

Want to check whether your AI-assisted CER process can show its work? Contact the Qmedify Regulatory Team.