The Guidance Cannot Outrun the Model

The Guidance Cannot Outrun the Model

FDA has authorized more than 1,450 AI- and ML-enabled medical devices, with nearly 300 clearances in 2025 alone barchart.com. ThinkSono’s DVT ultrasound guidance software just joined that list venousnews.com. Each clearance generates its own evidentiary record, its own performance claims, its own post-market drift profile. No single guidance document can hold all of that steady, and the agency knows it.

That is the quiet admission inside a warning circulating in PharmaVoice: “The fastest guidance that you could put out would still probably take a year. A year in the age of AI is very slow” pharmavoice.com. The observation isn’t a complaint about bureaucratic sluggishness. It is a structural fact about rulemaking timelines colliding with model iteration cycles that now run in weeks. The same mismatch is visible in Europe, where a Nature commentary is calling for dedicated Centres of Excellence in Regulatory Science to generate the concepts, evidence base, and horizon-scanning capacity that fit-for-purpose medical AI regulation requires, because the current apparatus wasn’t built for this cadence nature.com.

For compliance and technology leaders in regulated life sciences, this changes the planning assumption. The default posture for the last decade was to track FDA and EMA guidance, align internal controls to what regulators published, and treat the guidance document as the ceiling and the floor of your obligation. That posture assumes regulatory output moves at a pace roughly matched to product change. It no longer does. Databricks’ own guidance to healthcare organizations building AI capability is telling in its wording: track “evolving” frameworks across FDA and the EU AI Act implementation timeline, not settled ones databricks.com.

The practical consequence is that governance has to be built to outlast the guidance gap, not to wait for it to close. An organization deploying AI in diagnostics, drug discovery, or trial operations needs an internal control framework, ISO 42001 is the most direct fit, that can absorb model updates, new clearances, and shifting risk classifications without depending on the next regulatory bulletin to tell it what’s acceptable. That means version-controlled model change logs, documented risk rationale for every update, and audit trails that would satisfy a regulator even in the absence of a specific rule addressing that exact scenario. The EU’s push for regulatory science centres and the FDA’s acknowledged one-year lag are both signals that the agencies expect industry to carry more of this weight internally, not less.

This is not a call to deprioritize regulatory tracking. It is a call to stop treating it as sufficient. The organizations that will clear audits smoothly over the next two years are the ones whose internal governance already assumes the guidance will always be a step behind the model, and who built controls accordingly rather than controls that simply mirror the last published rule.

The gap between AI velocity and rulemaking velocity is not closing. Plan for the distance, not the bridge.

---

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 argument—that governance must be built to outlast regulatory lag—is coherent and logically structured, but the leap from ‘guidance takes a year’ to ‘agencies expect industry to carry more wei
Source & Claim Verification	Qwen · local	cleared. Most factual claims are supported by citations, but a few lines lack specific references, such as the claim about the number of AI- and ML-enabled medical devices cleared in 2025.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects the regulatory lag and the need for internal governance (e.g., ISO 42001) but does not explicitly detail compliance requirements under EU AI Act, FDA, or MDR/IVDR.
Technical Accuracy	Llama	cleared. The article accurately captures the challenge of regulatory frameworks keeping pace with AI advancements in life sciences, citing relevant sources and industry concerns.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and addresses potential vendor hype by framing the issue as a structural challenge rather than a product-specific problem, and it offers a clear counter-narrative t
Novelty & Non-Duplication	Grok	held. The core claim that AI model velocity outruns FDA/EMA guidance is commodity wire copy, and the piece sits too close to multiple already-listed back-catalogue items on the same guidance-gap thesis with
Validation	DeepSeek	cleared. Multiple cited sources, including FDA commentary and industry analysis, factually support the core claim that official guidance publication timelines are structurally slower than AI model iteration cy

Sources cited: 14. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.