Life Sciences / Regulatory Brief 🧬

📌 Navigate

📊 Exec Summary

The FDA executed a coordinated January package across adjacent windows — W01 (Jan 6: CDS and General Wellness final guidances), W01 (Jan 14: FDA x EMA joint AI-in-drug-development principles), and now (W04: cleared device baseline published; 2026 guidance agenda formalized). Together these moves represent the most significant deregulatory reset for AI-enabled digital health products since 21st Century Cures. This brief covers the W04 context window but treats all January FDA actions as a single package for clarity.

Five things moved in the January package:

The pattern: FDA is simultaneously pulling back pre-market requirements for lower-risk digital health products (CDS, wellness) and building the international alignment infrastructure (FDA × EMA) that will govern higher-risk AI in future binding guidance — a two-track reset with significant implications for SaMD and diagnostic AI builders.

1️⃣ FDA final CDS Software guidance

TL;DR: FDA finalized updated Clinical Decision Support Software guidance on January 6, 2026, introducing enforcement discretion for single-output AI recommendations where a clinician can independently review the basis — and reissued it on January 29, 2026 — materially expanding what SaMD builders can ship without 510(k) clearance.

What happened

• Final guidance issued January 6, 2026, then reissued January 29, 2026 (supersedes 2022 version); issued without prior public comment period
• Core change: enforcement discretion extended to CDS functions providing "a single output or recommendation" where "only a single option is clinically appropriate" — provided the HCP can "independently review the basis"
• New product categories now potentially non-device: cardiovascular event risk prediction from patient data, cognitive impairment treatment plan generation, radiologist finding summaries with diagnostic recommendations (if HCP review is required)
• Temporal scope expanded: time-critical decision-support functions no longer automatically excluded from non-device status
• AI/ML-specific: FDA evaluation focuses on HCP reviewability regardless of whether AI generated the recommendation — no AI-specific test

📊 Key facts

🔗 Primary source → FDA "Cuts Red Tape" on CDS Software (Arnold & Porter)

🔍 The non-obvious point

"HCP can independently review the basis" is an interpretable standard with no bright line — this creates both opportunity and risk for SaMD builders.

• Products that previously required 510(k) clearance (cardiovascular risk models with a single recommendation output, radiology summary tools) may now qualify for non-device status — but only if you can document that HCP review is structurally required, not just recommended
• The "independently review" language will be stress-tested by FDA in enforcement actions; early decisions (likely 2027) will define the practical boundary
• Watch for FDA to require transparency about model inputs and logic as a condition of meeting the "reviewability" standard — this is where explainability becomes a compliance requirement, not just a design preference

👀 What to watch

• First FDA enforcement action or warning letter citing the new CDS standard (expected 12–18 months post-guidance); comment deadline on any companion DSF guidance update

2️⃣ FDA General Wellness guidance update

TL;DR: FDA's revised General Wellness: Policy for Low Risk Devices guidance (January 2026) substantially expands what wearable devices can measure and display without triggering device regulation — explicitly covering optical blood pressure estimation, heart rate variability, sleep quality, stress, and recovery metrics.

What happened

• Final guidance issued January 6, 2026 (supersedes 2019 version); announced by Commissioner Makary at CES
• Expanded qualifying wellness product features: non-invasive optical sensing for heart rate and blood pressure, validated clinical value metrics, general notifications when outputs fall outside normal ranges, trend display for sleep quality, activity, stress, and recovery
• Wearable blood pressure monitoring (optical sensing method) explicitly permitted as a general wellness function — a reversal of the prior stance that restricted blood pressure monitoring to cleared devices
• Makary framed as FDA needing to "adapt with the times"; previewed a forthcoming "new regulatory framework for AI" with a "risk-based" and "deregulatory direction"
• Guidance also signals FDA will eliminate at least 50% of software and digital health guidances for clarity

📊 Key facts

🔗 Primary source → FDA Adapts with the Times on Digital Health (Ropes & Gray)

🔍 The non-obvious point

The blood pressure reversal is the operational headline for device builders — but the "notifications when outside normal range" language creates a new product design constraint.

• Wearables can now display and notify on blood pressure trends without clearance — but the guidance language implies outputs must remain "general wellness" in framing, not clinical diagnostic language
• This creates a labeling design test: the same optical BP sensor on the same wrist can be wellness (cleared for general use) or diagnostic (requiring 510k) depending entirely on how outputs are framed in UI/UX
• Regulatory strategy for wearable builders is now primarily a product language and UX exercise, not a hardware or algorithm exercise

👀 What to watch

• FDA Device Software Functions guidance update (promised in the January package); enforcement action on blood pressure wearables that cross the general wellness / clinical diagnostic line

3️⃣ FDA × EMA joint AI-in-drug-development principles

TL;DR: FDA and EMA co-published 10 high-level "Guiding Principles of Good AI Practice in Drug Development" on January 14, 2026 — the first major transatlantic regulatory alignment document on AI in the drug development lifecycle, covering evidence generation through manufacturing.

What happened

• Published jointly January 14, 2026; non-binding but explicitly framed as "foundation for future guidance, standards and harmonized regulatory expectations"
• Scope: AI systems used to generate or analyze evidence in nonclinical, clinical, post-marketing, and manufacturing phases for drugs and biologics
• Four enumerated principle areas: human-centric ethical design, risk-based development and performance assessment, data governance and cybersecurity, data quality and lifecycle management — six additional principles not individually enumerated in public summaries
• Both agencies signal formal binding guidance will follow; no specific timeline given
• Structured as pre-competitive alignment: intended to inform industry before requirements are set

📊 Key facts

🔗 Primary source → FDA and EMA Provide Guiding Principles for AI in Drug Development (McGuireWoods)

🔍 The non-obvious point

For drug developers using AI in evidence generation, this document is a preview of submissions requirements — not a current requirement.

• Any company submitting IND or NDA materials that include AI-generated datasets or AI-assisted clinical trial design should begin aligning to these principles now — regulators will expect to see evidence of this thinking in Q submissions and Pre-NDA meetings within 12–18 months
• The transatlantic alignment is strategically significant: a single development program can now anticipate converged FDA and EMA expectations on AI methodology, rather than managing two separate regulatory philosophies
• "Data quality and lifecycle management" as a principle category directly implicates AI system versioning, model change logs, and training data provenance documentation — this will become an inspection expectation

👀 What to watch

• FDA Q-Sub meetings and NDA review feedback where these principles are cited; EMA scientific advice requests referencing the joint document; anticipated formal guidance in 2026–2027 timeframe

4️⃣ FDA 2026 device guidance agenda

TL;DR: Hogan Lovells published a synthesis of FDA's stated CDRH guidance agenda for 2026 in January, identifying three transformational items: a new risk-based AI framework for medical devices, 50% reduction in software and digital health guidances, and update to Device Software Functions guidance.

What happened

• Hogan Lovells published CDRH agenda analysis, January 15, 2026
• FDA committed to: (1) new risk-based AI regulatory framework for medical devices — "smarter and more forward-thinking" per Makary; (2) eliminate at least 50% of existing software/digital health guidance documents for clarity; (3) update Device Software Functions and Mobile Medical Applications guidance (last significantly updated 2019)
• Framing: deregulatory in posture, but structured deregulation — the AI framework implies new standards replacing old ones, not elimination of standards
• Signals major structural reset for how 510(k) clearance will work for AI-enabled devices in 2026–2027

📊 Key facts

🔗 Primary source → FDA Device Guidance Agenda: What to Watch in 2026 (Hogan Lovells)

🔍 The non-obvious point

The 50% guidance reduction is a risk event, not just simplification — it will create interpretive gaps that enforcement actions will fill.

• When FDA removes existing guidances without replacing them, it does not eliminate the underlying statutory requirements — it removes the agency's own interpretive framework, leaving manufacturers to determine compliance on their own
• The new risk-based AI framework will define the clearance pathway for the next generation of AI diagnostic and SaMD products; early drafts and comment periods will be the most important regulatory documents of 2026 for device builders
• Internal FDA Agentic AI Challenge (launched January 2026) is a signal that FDA itself is piloting AI in review operations — watch for downstream changes to submission format and interaction expectations

👀 What to watch

• Draft risk-based AI device framework publication (most important document of 2026 for CDRH); Device Software Functions guidance update comment period; which existing guidances are consolidated or eliminated

5️⃣ AI-enabled device clearance baseline

TL;DR: By early 2026, FDA had authorized over 1,350 AI-enabled devices — double the 2022 count — with 295 clearances in 2025 alone and a median review time of 142 days, establishing the baseline against which January's deregulatory moves will be measured.

What happened

• FDA's AI-enabled device database stands at 1,350+ authorized products as of early 2026 (up from ~670 in 2022, ~950 as of August 2024)
• 295 AI/ML device clearances in 2025 — a single-year record; median clearance time 142 days, average 150 days
• 96.4% of AI-enabled devices cleared via 510(k) pathway; PMA pathway rarely used
• Majority in imaging AI (radiology, pathology, ophthalmology), clinical decision support, and vital signs monitoring
• January guidance changes may reduce the denominator (fewer products will require clearance) even as the rate of submissions continues to grow

📊 Key facts

🔗 Primary source → 2025 Year in Review: AI/ML Medical Device 510(k) Clearances (Innolitics)

🔍 The non-obvious point

The January deregulatory moves will shrink the cleared device count artificially — creating a potential false signal of slower adoption.

• Products that previously required 510(k) clearance but now qualify as non-device CDS or wellness tools will no longer appear in the FDA authorization database — a useful leading indicator may become harder to read
• The 96.4% 510(k) concentration means any change to the 510(k) predicate system has outsized impact on AI device clearance timelines; the new AI framework may disrupt predicate chains
• Builders planning 510(k) submissions in 2026 should model two scenarios: current pathway and post-framework pathway (once the new risk-based AI framework drops mid-year)

👀 What to watch

• Monthly FDA AI device database additions post-January guidance; 510(k) predicate challenge rate for AI devices in 2026 first half

📊 The pattern

FDA executed a coordinated January package: two final guidances reducing pre-market requirements (CDS, General Wellness), a joint EMA principles document setting the trajectory for binding drug-development rules, and a stated agenda that promises a new risk-based AI device framework mid-year. The direction is consistent — deregulate the known, build new infrastructure for the unknown — and the pace is faster than most device builders' regulatory calendars anticipated. The 1,350-device clearance baseline is the before-state; the post-January environment will look meaningfully different for SaMD builders, wearable manufacturers, and CDS platform developers.

👀 Watchlist

📎 Sources

Sources of truth

Also consider reading