Authorization volume is no longer the bottleneck.
FDA's AI-enabled device list makes the landscape visible enough to map by use case, specialty, and regulatory pathway.
Digital HealthClinical AIDigital biomarkersDeployment drift
Digital Health
As of June 2026, digital health has moved from model demonstration into lifecycle proof. FDA-authorized AI devices, sensor-derived digital health technologies, AI-supported regulatory evidence, and adaptive clinical tools are all expanding. The hard question is which model or biomarker changes a clinical or research decision under deployment conditions.
Domain research lens
This page tracks where AI is authorized, where prospective validation is still thin, which digital biomarkers can survive drift, and when a model output becomes decision-grade under deployment conditions.
June 2026 field state
A field shifting from prediction demos to monitored clinical utility.
The strongest work now asks whether an AI system or digital biomarker improves decisions across patients, sites, time, workflow, and regulatory context.
Clinical AI, digital biomarkers, computational biology, prospective validation, deployment drift, sensor-derived measures, and decision utility.
The domain thesis: medical AI value is not the score at launch. It is whether the model remains useful, safe, fair, and decision-changing after the world around it moves.
Lifecycle rules are becoming concrete.
Predetermined change-control plans let adaptive AI be reviewed around bounded, monitored updates.
Generalization is still the hard evidence layer.
Many systems clear narrow tasks before they prove prospective utility, workflow fit, fairness, drift control, and decision impact.
Recent research signals
The 2025-2026 update is a decision-gate wave.
Each signal below starts from the field: what changed, why it matters, and which research or buyer decision becomes more testable.
AI-enabled devices are now a regulated landscape.
FDA maintains a public list of AI-enabled medical devices authorized for marketing in the United States.
Why it matters
The buyer problem shifts from whether AI can be authorized to which devices have enough clinical validation, monitoring, and workflow evidence.
Decision implication
Changes diligence from feature review to lifecycle-risk review.
Predetermined change-control plans became central to adaptive AI.
FDA guidance provides recommendations for PCCPs tailored to AI-enabled devices.
Why it matters
Adaptive models are only useful if updates are governed, bounded, and monitored.
Decision implication
Changes whether a buyer trusts a static model, monitored model, or continuously updated model.
AI used for regulatory evidence needs context-of-use credibility.
FDA's drug and biologic AI guidance uses a risk-based credibility framework tied to context of use.
Why it matters
AI biology claims are weaker when the model is judged apart from the decision it supports.
Decision implication
Changes whether a model is used for hypothesis generation, candidate triage, or regulatory-grade evidence.
Digital biomarkers are becoming an inspectable device landscape.
FDA's sensor-based DHT list identifies authorized medical devices that use sensor-derived measures.
Why it matters
DHT value depends on analytic validity, clinical meaning, endpoint acceptance, and real-world signal quality.
Decision implication
Changes whether a measure is a convenience feature, endpoint candidate, or decision-grade biomarker.
What leading groups are working on
The current edge is a set of validation programs, not a single breakthrough story.
These representative programs and research fronts frame the active field before the page maps Zemi Dossiers into it.
Regulated AI devices
The FDA AI-enabled device list becomes a landscape map
By June 2026 the field is crowded enough that diligence must sort authorized use case, specialty, regulatory pathway, clinical validation, workflow integration, and monitoring obligations.
Adaptive AIPredetermined change-control plans as the lifecycle gate
PCCPs make model updating inspectable, but only if the manufacturer defines bounded changes, validation data, risk controls, and postmarket monitoring.
Drug development AIContext-of-use credibility for regulatory evidence
AI-generated drug or biologic evidence now needs a risk-based credibility argument tied to the exact decision the model supports.
Digital biomarkersSensor-derived measures become an authorized-device surface
Wearables, passive sensors, ECG, imaging, gait, speech, and retinal measures are valuable only when analytic validity, clinical meaning, and endpoint acceptance are clear.
Deployment evidenceReal-world performance, drift, and human factors
The hard innovation is maintaining usefulness after launch, when populations, workflows, devices, labels, and users change.
Decision gates
What must be true before a buyer should build, fund, partner, monitor, avoid, or run the next study.
These are field-level gates first. The dossier library appears later as the set of existing Zemi products that can help investigate them.
Prospective utility
Does the tool improve a decision or outcome against current standard?
Generalization
Does performance survive new sites, devices, populations, disease mix, and workflow context?
Drift control
What breaks when distributions, users, labels, or model versions change?
Endpoint acceptance
Can the AI or digital biomarker support a decision regulators, clinicians, or sponsors will accept?
Human factors
Does the workflow reduce burden without creating automation bias or silent error?
Evidence tiering
Which claims are known, inferred, hypothetical, or unsupported?
Zemi Dossiers in this domain
The dossiers sit where new research creates hard buyer decisions.
Each dossier card uses stats from the actual research report manifest and Evidence & Decision Workbook, including pages, workbook sheets, evidence/source rows, claim rows, power rows, and decision instruments where present.
Primary Zemi Dossier
AI Biology Drug Discovery
Separates static generation wins from the harder validation problem: whether AI can predict dynamic biological response under prospective tests.
Why it belongs here
Which AI-biology programs deserve validation spend now, and what experiment would show whether the model changes a real discovery decision?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Primary Zemi Dossier
AI Clinical Validation & Digital Biomarkers
Turns AI and digital biomarkers into validation-gate questions: prospective utility, generalization, drift, fairness, and endpoint acceptance.
Why it belongs here
Which AI or digital-biomarker claims are ready for prospective validation, and which are still retrospective performance stories?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Engineering Memory / Engrams
Resolves the Lost-vs-Locked diagnostic gate before choosing re-access, editing, stabilization, reconstruction, or avoidance.
Why it belongs here
Which memory programs are actually testable, and what state classifier must exist before choosing an intervention?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Neurodegeneration Convergence
Uses convergent detection to subtype disease, while requiring sign-labile biology and therapeutic-window logic before treatment spend.
Why it belongs here
Which neurodegeneration programs are using convergence for detection while still measuring the divergent biology needed for treatment decisions?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
BCI / BSI Long-Term Stability
Diagnoses whether BCI/BSI decline is recoverable code drift or irreversible source loss before buyers overspend on the wrong layer.
Why it belongs here
Is performance decay a software problem, a biology/materials problem, or a mixed failure that requires a different validation plan?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Bioelectronic Neuro-Immune Closed Loop
Scores indication readiness by biomarker validity, sensing fidelity, decoding generalizability, circuit match, and substitution economics.
Why it belongs here
Which indications have enough biomarker, sensing, decoding, and reimbursement logic to justify a closed-loop bioelectronic program?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
MRD-Guided Neoantigen Vaccines
Links MRD timing, antigen selection, vaccine manufacture, immune response, and recurrence endpoints into a recurrence-prevention decision gate.
Why it belongs here
Which MRD-guided vaccine strategies have enough timing, manufacturing, immune-response, and endpoint logic to justify next-step validation?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
RNA Editing & the Permanence Spectrum
Positions RNA editing on a permanence spectrum so buyers do not confuse reversibility with durability or safety by default.
Why it belongs here
Which RNA-editing programs should rent correction, extend correction, or avoid RNA-level strategy based on permanence needs?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Genetic Cardiomyopathy Precision Therapies
Uses molecular mechanism rather than gene label to route programs toward addition, knockdown, editing, or avoidance logic.
Why it belongs here
Which genetic cardiomyopathy programs match mechanism, modality, delivery, safety window, and evidentiary standard well enough to advance?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
AI Multi-Omics Variant-to-Rescue
Routes variant interpretation toward rescue experiments instead of stopping at association, prediction, or annotation confidence.
Why it belongs here
Which variant-to-function programs have enough evidence to justify rescue experiments, and what readout would falsify the rescue thesis?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Precision AMR Countermeasures
Uses mutational supply, diagnostic timing, pathogen burden, and combination logic to assess durability against resistance escape.
Why it belongs here
Which AMR countermeasures reduce escape paths enough to justify development, deployment, or monitoring priority?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Organ-on-Chip / NAM Qualification
Routes a proposed context of use to its binding qualification gate before buyers spend on biological completeness.
Why it belongs here
Which context of use is the chip actually qualified to support, and what reproducibility or anchor-transfer evidence is still missing?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Adjacent / cross-domain
Ovarian Aging Multi-Clock
Separates endocrine, follicular, mitochondrial, stromal, and functional clocks before buyers infer benefit from measurement movement.
Why it belongs here
Which ovarian-aging readouts represent functional benefit, and which only move a clock without changing the decision?
Pairs the research report with workbook evidence rows, claim discipline, decision instruments, power calculations, and next-study surfaces.
Watchlist
Signals that would change the map in 2026-2028.
These are the developments most likely to change the field map, evidence posture, or next-study priorities.
AI devices
FDA AI list growth, PCCP use, and recalls
Volume alone is not proof. The next useful map separates authorized tasks, model-change permissions, adverse events, and clinical utility.
Clinical utilityProspective studies reporting utility rather than accuracy alone
The most important evidence is whether AI changes care, discovery, trial design, or operational burden under deployment conditions.
Digital endpointsEndpoint qualification and sensor-based DHT authorizations
Sensor measures need a path from clean signal to accepted endpoint or decision-grade enrichment measure.
AI biologyVariant-to-rescue and target-discovery experiments
AI biology earns value when predictions close into prospective perturbation, rescue, or candidate-selection experiments.
Field-note discipline
Current-state pages need a source-aware update layer.
The public page stays readable, but the underlying domain model tracks source-linked developments that change evidence posture, buyer decisions, or next-study priorities.
June 2026 / AI devices
The domain problem has moved from authorization scarcity to lifecycle quality.
The question is which authorized tools can prove utility, fairness, drift control, and workflow fit after launch.
2025 / PCCP Adaptive AI now has an inspectable regulatory mechanism.Change-control plans make continuous improvement credible only when the update envelope and validation plan are concrete.
2025-2026 / DHT Digital biomarkers need endpoint meaning, not just continuous data.Sensor-derived measures must show why the signal matters clinically or for trial decisions.
From domain signal to Zemi Dossier
Request access to inspect the full research report, Evidence & Decision Workbook, power calculations, and release-audit surfaces behind each decision package.