Artificial intelligence is transforming healthcare. But faster recommendations do not necessarily mean better clinical decisions when the underlying context is incomplete.
A perspective on clinical continuity, accountability, and patient-controlled information sharing.
Medicine rarely unfolds within a single institution, a single professional relationship, or a stable patient situation. Yet an AI system may be asked to interpret the fragments it can access as though they represented the whole.
The result can be more dangerous than an obvious gap. A coherent-looking medical record may conceal events that occurred elsewhere, decisions made by other clinicians, and changes in the patient’s circumstances. If those omissions are not visible, neither the clinician nor the AI system necessarily knows what to question.
Artificial intelligence is accelerating clinical decision-making faster than healthcare systems are improving their ability to preserve the context behind those decisions.
01
A patient may see several primary care physicians, specialists in the same discipline, or covering clinicians over time. Each may make reasonable decisions based on a different subset of the patient’s history.
What may be missing: prior concerns, rejected hypotheses, medication changes, unresolved questions, or reasons another clinician chose a different course.
02
Clinics, hospitals, diagnostic centers, and independent practices may deliver care sequentially without exchanging the relevant information. Institutional boundaries and commercial competition do not disappear simply because the patient moves between them.
What may be missing: investigations, treatments, adverse events, clinical discussions, and follow-up plans held in another organization’s systems.
03
Patients may stop taking a medication, change their dosage, seek treatment elsewhere, use over-the-counter products, miss follow-up, or experience new social and functional constraints. These changes can materially alter the significance of earlier information.
What may be missing: the reason a treatment was not followed, a new exposure, a change in adherence, or a developing problem that was never communicated or recorded.
04
A recommendation may be internally consistent with the information presented and still be inappropriate because the patient’s context has changed. Faster processing does not compensate for an absent fact that would have changed the decision.
What may be missing: a decisive contraindication, a new diagnosis, a revised preference, a recent test, or a previously documented concern that was not shared.
An AI-generated answer can appear comprehensive even when the underlying data are incomplete. The system may not be able to recognize that a relevant event, conversation, or change is absent from its inputs.
That creates a particularly consequential clinical blind spot: the assumption that everything needed for a decision has already been considered. A rapidly produced recommendation can accelerate action, reinforce unjustified confidence, and reduce the opportunity to discover missing context before treatment begins.
Information dispersed or undocumented
Absence is not recognized
Apparent completeness
When action outruns verification
This is a risk mechanism, not a claim that every AI system behaves this way or that human clinicians are immune to the same failure. The practical concern is the scale and speed at which incompletely grounded decisions could be produced and acted upon.
A patient receives a treatment recommendation at one clinic. Later, a different organization adjusts the medication after an adverse reaction. The patient then stops taking it, but that decision is not communicated to the first clinic. Months later, a clinician uses an AI assistant to review the first clinic’s records.
The assistant may generate a plausible next step from the documented history. It may have no reliable signal that the medication was changed elsewhere, that the patient stopped it, or why. The missing information is not merely an empty field; it is an unrecognized change in the clinical situation.
Hypothetical example for explaining the information problem; not a statement about a particular AI product, patient, or medical recommendation.
Clinical continuity depends on more than transferring a final note. It can require identifying the participants, the sequence of exchanges, what information was available at different points, what was asked or reported, and what permissions govern subsequent access.
Understand when observations, questions, responses, and changes occurred across an episode of care.
Know who communicated information, and distinguish recorded facts from assumptions or later interpretations.
Respect granular sharing choices and revocation, even when an earlier participant previously had access.
No platform can reconstruct undocumented events or guarantee a complete record of a patient’s changing circumstances. Preserving documented context reduces avoidable losses; it does not eliminate clinical uncertainty.
Through InTouch, PortableEHR supports persistent, chronological communication threads within episodes of care, identifiable participants, timestamped exchanges, and patient-controlled permissions. Authorized clinicians and collaborators can review the documented exchanges that remain available to them rather than relying exclusively on isolated end-point summaries.
Patients can selectively authorize and revoke access. If information becomes restricted, participants must not be assumed to retain continuing visibility simply because they were previously involved. These safeguards are central to the architecture, not optional extras.
What this contributes: a more traceable and patient-governed foundation for communication and continuity among participating people and organizations.
What it does not claim: to capture every clinical thought, connect automatically to every healthcare institution, infer undocumented patient behavior, guarantee correct decisions, or currently provide AI-driven diagnosis.
We should not confuse the ability to generate a clinical recommendation with the ability to justify a clinical decision.
A better AI system should not only analyze what healthcare already knows. It should also be able to identify consequential gaps—and, when appropriately authorized, seek the information that healthcare still needs to know.
Clinical information is not confined to institutional systems. Relevant changes may be known only to a patient, a guardian or caregiver, an outside clinician, or potentially a validated connected device. When an AI-assisted workflow encounters uncertainty, a future integration could help it formulate specific questions and request new information rather than silently treating an incomplete record as complete.
With the patient’s consent and the care team’s authorization, an appropriately integrated AI assistant could identify unanswered questions—for example, recent hydration, medication changes, symptoms, or environmental exposures—and initiate a controlled request to the patient or authorized representative. Documented responses could then inform a clinician’s reassessment before a non-urgent therapeutic decision.
Within an approved research protocol, an authorized research team could potentially use a similar mechanism to request targeted follow-up information from consenting participants: treatment experience, adverse events, patient-reported outcomes, or other relevant observations. This would require study-specific permissions, privacy safeguards, appropriate ethical and regulatory approvals, and validated data-collection procedures.
The potential loop: recognize uncertainty → identify a relevant question → verify authority and consent → contact the appropriate person or approved source → document the response and its provenance → reassess the original hypothesis.
AI should not only analyze what healthcare knows. It should help discover what healthcare still needs to know.
Why PortableEHR may be relevant: InTouch already provides foundational capabilities for authenticated, bidirectional communication, patient-controlled access, revocation, and timestamped exchanges. With targeted integrations and the necessary resources, these could support a future authorized recontact pathway for an AI system acting on behalf of a care team or research program.
Development status: This is a prospective integration concept, not a currently available autonomous AI feature or a claim of clinical or research validation. Any implementation would need clearly defined roles and consent, secure integration, clinical or study oversight, appropriate validation, and safeguards for urgent cases. A response from a patient or device is evidence to assess—not automatic proof of its accuracy. Revocation and downstream use of already collected research data would be governed by the applicable consent terms and legal requirements.
No AI system should be the sole authority responsible for generating, validating, and investigating its own conclusions. Healthcare organizations need the ability to examine decisions and investigate unexpected behavior independently of any single AI provider.
Independent verification is not a rejection of AI. It is a prerequisite for accountable AI.
The July 2026 OpenAI–Hugging Face cybersecurity incident illustrated why this matters. Autonomous AI agents operating in a security evaluation breached intended boundaries; Hugging Face subsequently used an alternative open-weight AI model as part of its investigation. The incident does not show that multiple models would have prevented the intrusion. It shows the value of retaining independent investigative capabilities.
Further reading: OpenAI incident account. The incident is a cybersecurity example, not a clinical event.
Provenance, recency, consent and missing information
Other sources, alternative models and auditable investigation
Clinical judgment, documented rationale and oversight
PortableEHR’s potential role is in the independent identity, consent, communication and traceability layer—not in providing model cybersecurity testing, autonomous diagnosis or a multi-model verification engine.
Clinical AI deployment should ask not only what a model can conclude from available information, but also what sources were consulted, what might be missing, how current the information is, and which decisions require independent verification.
Structured communication and traceable context can support that discipline. Any future use of InTouch context by clinical or research AI—including targeted recontact—would require appropriate integrations, purpose-specific permissions, consent and access controls, validation, and clinical or research oversight. Availability of a communication history does not itself establish its completeness or fitness for a particular clinical decision.
Explore how patient-controlled communication and clinical continuity could strengthen your organization’s workflows—and inform a responsible approach to AI in medicine.
PortableEHR supports clinical communication and documented context. It does not replace professional judgment or provide an autonomous clinical decision service.
