AI Governance's Real Fault Line Is Classification, Not Geography

Multinational life sciences companies have spent two years building AI governance around a shared assumption: that ISO 42001 certification, EU AI Act compliance, and FDA engagement form a compatible stack, with local rules layered on top. The harder problem isn’t that this assumption is wrong in general. It’s that regulators in the US, EU, and China are drawing different lines around what counts as a regulated AI function in the first place, and those lines are moving independently.

China’s regulators have expanded oversight across the life sciences value chain, from target discovery and clinical trial design through patient recruitment, imaging analysis, and digital therapeutics, with new compliance obligations arriving faster than most global programs were built to absorb HLC. That scope now overlaps directly with where AI adoption is deepest globally. Generative biology tools are already reshaping early discovery workflows Forbes, and AI increasingly touches virtual control groups and clinical trial operations industry-wide Technology Networks.

The classification problem, not just the geography problem

The sharper issue is that classification decisions inside each jurisdiction are unsettled on their own terms, before you even compare across borders. In the UK, the medicines regulator has had to weigh in directly on whether AI scribes qualify as medical devices, a determination that changes which regulatory pathway applies to a tool already in clinical use MHRA coverage. In the US, the FDA’s draft MDUFA VI commitment letter signals how device user fee categories are being reworked specifically to account for AI-enabled digital health products, a structural shift in how the agency will treat these tools at intake MDUFA VI analysis. Meanwhile routine clinical health data is now outperforming curated trial data for training AI models, and FDA guidance has not caught up to that shift in what counts as acceptable training data provenance Clinical Trial Vanguard. These are three separate classification fights, each unresolved, each moving on its own clock.

Europe’s response has been to build granular tracking rather than resolve the classification question outright. The EU Clinical Trials Regulation dashboard now compiles trial application decisions with increasing jurisdiction-specific detail Jones Day, which gives regulators visibility without forcing convergence on definitions.

What this means for the governance build

A sponsor running AI-assisted recruitment or imaging analysis across a multi-region trial isn’t just facing three regulatory postures. It’s facing three unresolved answers to the question of whether that function is a device, a data source, or neither, and those answers are shifting inside each jurisdiction before you even get to comparing them.

The practical fix isn’t forking ISO 42001 or the EU AI Act into a China-specific branch. It’s building a live classification map, function by function, jurisdiction by jurisdiction, tracking MDUFA fee categories, MHRA device determinations, and China’s expanding scope as moving inputs rather than settled facts. A framework certified against yesterday’s classification lines is not compliance. It’s a snapshot.

Companies that treat classification as a one-time mapping exercise are betting that the lines hold still long enough to matter. Given the pace on all three fronts, that bet is already behind schedule.

---

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 central thesis—that classification instability within jurisdictions matters more than geographic divergence—is coherent and defensible, but the argument relies heavily on asserting that classifica
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but some citations are redundant and could be streamlined for clarity.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects the evolving and jurisdiction-specific classification challenges under ISO 42001, EU AI Act, FDA, and MDR/IVDR, though it could explicitly map its claims to specific c
Technical Accuracy	Llama	cleared. The article accurately reflects the current regulatory landscape and challenges in AI governance for life sciences, citing relevant sources and technical details.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and supports its core counterargument regarding classification, with minimal vendor hype.
Novelty & Non-Duplication	Grok	held. The ‘classification not geography’ frame is only a modest rephrase of extensively covered SaMD/device-boundary fights and jurisdictional fragmentation already standard on the wire and in life-sciences
Validation	DeepSeek	cleared. The central claim that regulatory classification of AI functions is dynamic and divergent across major jurisdictions is strongly supported by cited evidence of active, unresolved determinations in the

Sources cited: 10. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.