The familiar picture of bio-digital sovereignty begins with intimate artefacts: DNA files, retinal scans, gait signatures, neural recordings, fertility data, clinical notes. The policy instinct is therefore straightforward. Protect the sensitive stuff, restrict access, require consent, and punish leaks. That agenda remains necessary, but by mid-2026 it is no longer sufficient. The decisive shift is from governing collection to governing interpretation.
What now matters most is inferential capacity: the ability of software, institutions and states to derive facts about a person’s body or mind from data that did not previously count as bodily in any obvious sense. A blood sample is sensitive, but so is a purchasing history once machine learning turns it into a proxy for pregnancy, depression or cognitive decline. A face image is sensitive, but so is a voice recording once it is used to estimate fatigue, stress, intoxication or neurological disease. The sovereign question is no longer merely who owns bodily data, but who may construct a body from traces.
From biological samples to biological models
This is not simply a technical refinement. It is a constitutional change in the information order around the human person. In the older model, institutions harvested a discrete bodily signal, stored it, and used it for a bounded purpose. In the newer one, they assemble a probabilistic portrait from dispersed traces: browsing patterns, keystroke dynamics, location histories, consumer records, speech cadence, wearable telemetry, CCTV footage, insurance forms and educational performance. The body becomes legible through correlation rather than direct measurement.
That matters because law and public intuition still treat sensitivity as a property of the input. Yet in practice sensitivity increasingly arises at the level of output. A dataset may look mundane in isolation, while the inference generated from it is deeply intimate and socially consequential. Existing categories such as health data, biometric data or special-category data under the GDPR remain important, but they do not fully resolve the political problem created when ordinary data are used to manufacture extraordinary claims about a person’s biological state.
Inference is the new site of power
There is a tendency to describe this as a privacy issue. It is one, but that description is too narrow. Inferences do not merely reveal; they govern. Once a system infers frailty, addiction risk, neurodivergence, pregnancy, pain, deception, agitation or reduced cognitive performance, institutions can sort, price, monitor, exclude or pre-empt. The practical force of the inference lies not in whether it is metaphysically true, but in whether it is operationally accepted by employers, insurers, schools, border agencies, welfare administrators or police.
This is where bio-digital sovereignty becomes materially political. Sovereignty concerns who gets to define the person in administratively actionable terms. If systems can derive bodily meanings from ambient data, then control over inference becomes a form of control over status. It determines who is seen as risky, impaired, fertile, stable, compliant or dangerous before any formal diagnosis exists and often without the subject’s knowledge.
A blood sample is sensitive, but so is a purchasing history once machine learning turns it into a proxy for pregnancy, depression or cognitive decline.
The law has begun to notice, but indirectly
The decisive shift is from governing collection to governing interpretation.
European regulation has started to approach this terrain, albeit through several separate doors rather than one coherent theory. The GDPR addresses profiling and special-category data, and the European Data Protection Board has increasingly wrestled with AI models and downstream processing. The EU AI Act goes further by treating certain biometric categorisation and emotion-recognition uses as especially hazardous or prohibited in particular contexts. UNESCO, the OECD, NIST and the WHO have each, in different vocabularies, stressed that AI systems can generate sensitive and harmful inferences well beyond the data categories that trigger older compliance habits.
Still, most governance frameworks remain more comfortable asking whether a system collected prohibited data than whether it derived prohibited meaning. The distinction is politically convenient because collection is easier to observe. Inference is murkier. It happens inside models, through feature engineering, score construction and proxy variables. Yet that murkiness is precisely why it deserves greater constitutional attention. Hidden interpretation is often more powerful than visible capture.
Why proxies are destabilising
Proxy inference undermines one of the classic bargains of privacy law: that citizens can manage exposure by deciding what not to disclose. In a world of dense correlation, refusal loses some of its force. One may never submit a mental health questionnaire, yet language patterns, sleep disruption, spending shifts and social withdrawal inferred from devices or services can produce a behavioural score suggesting depression. One may never hand over a fertility record, yet app use, pharmacy purchases and mobility changes can feed a model that predicts pregnancy likelihood.
That does not mean privacy is obsolete. It means privacy without inferential restraint is incomplete. The old promise was selective opacity. The new reality is involuntary legibility. Institutions no longer need direct access to the body if they can reconstruct enough of it from nearby signals.
The weakest point in current safeguards
Consent is particularly ill-suited to this environment. It cannot plausibly cover all future inferences derivable from apparently routine data, especially when even system designers may not anticipate every emergent correlation. Notice also breaks down. To inform someone meaningfully, one would need to explain not only what data are collected but what biological or mental attributes might later be inferred, by whom, with what confidence, and for which institutional decisions. Most privacy notices cannot do this honestly because the inferential ecosystem is too dynamic and too opaque.
Purpose limitation remains valuable, but purpose statements are often abstract enough to accommodate broad analytics. Security obligations help against theft, not against authorised over-interpretation. Even rights to access or correction have limits where the real issue is not a factual error in a raw record but a contestable statistical assertion about one’s future body, behaviour or capacity.
Neurodata is only the leading edge
The rise of consumer neurotechnology has rightly drawn attention to neural signals, cognitive liberty and mental privacy. But focusing too narrowly on explicit neurodata risks missing the wider structural issue. Many institutions will not need direct brain-computer interfaces to govern mental states. They will use speech, attention metrics, eye movements, interaction rhythms and other behavioural traces as stand-ins for concentration, emotional regulation, fatigue or susceptibility.
A blood sample is sensitive, but so is a purchasing history once machine learning turns it into a proxy for pregnancy, depression or cognitive decline.
The same pattern extends across the biological spectrum. So-called biodata are no longer confined to laboratory outputs or medical records. Biology is becoming computationally ambient. The frontier problem is not merely protecting novel sensors, but preventing routine environments from being converted into speculative diagnostic systems.
The sovereign question is no longer merely who owns bodily data, but who may construct a body from traces.
Administrative appetite will outpace scientific certainty
One reason this matters now is that institutions have incentives to operationalise weak science. Administrators do not wait for epistemic closure. If a model appears useful for screening applicants, allocating benefits, triaging inspections or identifying high-risk passengers, it may be deployed long before the underlying biological inference is robust across populations and contexts. The history of risk scoring in other domains shows how quickly opaque tools can acquire bureaucratic authority.
For bio-digital sovereignty, the danger is not only false positives. It is category inflation. Once systems produce scores for stress, attention, resilience, pain or trustworthiness, these constructs can migrate from research hypotheses into practical labels embedded in platforms and procedures. The state-like function is then exercised by a hybrid assemblage of software, procurement standards and institutional habits rather than by explicit legislation.
What a more serious doctrine would look like
A stronger doctrine of bio-digital sovereignty would start by treating certain inferences as regulated objects in their own right, irrespective of whether the source data were conventionally sensitive. The central question would be: what claims about bodies and minds may not be generated, sold, shared or used for decisions without a heightened legal basis and public justification?
This would shift governance towards use-based and output-based controls. Some inferences should simply be off-limits in specific settings: emotional state estimation in workplaces and classrooms, for instance, or predictive health and cognitive profiling for access to essential services. Others might be permitted only under strict conditions of necessity, independent validation, contestability and human review. The point is not to freeze research; it is to prevent the casual conversion of correlation into administrative fact.
Public institutions need a higher bar than private actors
States and quasi-public bodies deserve particular scrutiny because their decisions are hard to evade. A citizen can sometimes leave a service or decline a device; it is far harder to exit a welfare office, a public school, a border crossing or a criminal justice system. When public institutions use inferential systems about bodies and minds, they engage core questions of due process, equality and democratic legitimacy.
The sovereign question is no longer merely who owns bodily data, but who may construct a body from traces.
The appropriate standard therefore cannot be mere technical accuracy. It must also include normative fitness. Even a moderately accurate model may be illegitimate if it relies on concepts that are too subjective, too manipulable or too laden with historical bias to anchor official action. Not every measurable pattern should become a governable category.
Scientific humility is a constitutional value
There is an awkward gap between what machine learning can correlate and what democratic institutions should be allowed to conclude. Bridging that gap requires more than auditing. It requires scientific humility built into law. The WHO’s work on AI in health and the OECD’s neurotechnology recommendation both point, in different ways, to the need for proportionality, human rights and caution where uncertainty is high and consequences are serious.
That principle should be made sharper. Systems that infer biological or mental states from indirect signals should face rebuttable scepticism, especially where they affect liberty, livelihood, education, insurance or healthcare access. The burden should sit with deployers to show not only model performance, but conceptual validity: that the thing being measured is real enough, stable enough and relevant enough to justify institutional action.
Democracy must govern the categories themselves
Much current debate assumes that categories such as stress, engagement, deception, impairment or risk already exist in a politically neutral form, awaiting better measurement. They do not. These are socially freighted constructs. To encode them is to choose a theory of the person. Deliberative democracy matters here not as procedural decoration but as a check on silent ontology-making by technical systems.
Citizens should not discover after the fact that schools classify attention through gaze, employers infer burnout from keyboard patterns, or border systems estimate nervousness from facial micro-movements. The issue is not transparency alone. It is whether those categories should be administratively usable at all. A democratic polity must reserve the right to declare some forms of computational interpretation incompatible with personal dignity and civic equality.
A sovereignty agenda for the age of inference
The coming settlement in bio-digital governance will not be secured by adding ever more data types to lists of sensitive information. That strategy will lag behind a world in which almost any digital exhaust can become a substrate for bodily inference. What is required is a more ambitious constitutional grammar: one that recognises inferential power as a primary site of domination.
Bio-digital sovereignty, in this view, is not a property of databases. It is a rule about the limits of authorised interpretation. It asks when institutions may transform traces into claims about health, cognition, emotion or future capacity, and when they may act on those claims. The societies that answer that question well will not be those that simply guard samples more carefully. They will be those that insist the human body cannot be endlessly reconstructed by code without public permission, legal restraint and democratic reason.
