Two Speeds of FDA AI Clearance

Two Speeds of FDA AI Clearance

Tempus just picked up its third FDA clearance for an AI cardiac tool, this time for detecting signs of pulmonary hypertension from a standard ECG, joining earlier clearances for atrial fibrillation and low ejection fraction detection (BioSpace, MobiHealthNews). Three clearances in a fairly short span for a similar model architecture confirms something regulatory teams should already be treating as settled. The 510(k) pathway works cleanly for AI that takes a defined input, produces a single bounded clinical signal, and can be validated against a predicate.

That is not where the oversight pressure sits. The FDA has opened a public comment period on how to regulate medical devices powered by generative AI, running through October, and has described it as one of the hardest oversight questions the agency currently faces (Digital Journal). This is not an isolated gap limited to text-generating clinical tools. The same structural problem is surfacing in adjacent device classes. Therapeutic VR platforms, now reaching real clinical deployment, face an analogous question: how do you build an ongoing safety record for an intervention that adapts to the patient in real time rather than producing one fixed output, which is why practitioners are calling for something closer to a continuous safety ledger than a one-time clearance (HIT Consultant). Bounded diagnostic AI and adaptive therapeutic AI are running into the same wall from different directions, and the FDA’s comment docket is the first attempt to build a wall-sized fix.

For compliance and regulatory leaders, this is the real decision point, not any single clearance. If your pipeline resembles Tempus’s model class, structured input, bounded clinical output, clear predicate, the 510(k) route is proven and the compliance lift is mostly validation rigor and post-market surveillance discipline. If your pipeline includes generative or adaptive components, whether a clinical documentation assistant, a synthesis engine, or a therapeutic tool that changes behavior session to session, you are building toward a regulatory target that does not exist yet in fixed form. The open comment period is the one window where your organization’s operational experience can shape that target rather than simply inherit it.

The transatlantic picture raises the stakes. European MedTech bodies are simultaneously pushing to revise the EU AI Act’s Digital Omnibus and MDR/IVDR frameworks, explicitly to prevent AI innovation from migrating to the US or Asia while regulatory sequencing gets sorted out (Healthcare.Digital). Any life sciences organization deploying generative or adaptive AI across US and EU markets is tracking two moving targets at once, on different timelines, with different risk taxonomies, and now a shared underlying question about what counts as a stable, clearable system at all.

The practical move is unchanged but sharper now. Classify your AI portfolio by output type, bounded versus adaptive or generative, and route governance resources accordingly. The bounded systems need clearance discipline. The adaptive and generative systems, whether diagnostic, therapeutic, or documentation-facing, need a seat at the table while the rules are still being written. Waiting for final guidance to engage means complying with a framework you had no hand in shaping.

---

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 bounded diagnostic AI and adaptive/generative AI face fundamentally different regulatory challenges—is coherent and well-supported, though the claim that Tempus’s three clearanc
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but the article could benefit from more specific details on the regulatory challenges and timelines for generative AI and adaptive systems.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects FDA, MDR/IVDR, and EU AI Act distinctions for bounded vs. adaptive/generative AI, but lacks explicit ISO 42001 alignment references.
Technical Accuracy	Llama	cleared. The article accurately describes the current FDA clearance process for AI medical devices and highlights the challenges posed by generative and adaptive AI, although it includes some tangential source
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively distinguishes between two types of AI clearance, directly addressing potential vendor hype by clarifying which AI models fit existing regulatory pathways versus those facing s
Novelty & Non-Duplication	Grok	cleared. The two-speed framing linking Tempus’s repeatable 510(k) clearances to the open genAI docket, adaptive VR safety-ledger gap, and EU pressure is a fresh multi-source synthesis rather than a wire rewrit
Validation	DeepSeek	cleared. The central claim that the FDA’s 510(k) pathway works cleanly for bounded AI diagnostics is validated by Tempus’s repeated clearances, while the claim about regulatory uncertainty for adaptive/generat

Sources cited: 11. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.