Category: Risk Classification & Tiering

Minimal Risk

Simply put

Minimal risk is a threshold used in human subjects research to describe activities where the chance and severity of harm are no greater than what people ordinarily face in daily life or during routine physical or psychological examinations. When research is judged to fall below this threshold, it may be eligible for less intensive ethical oversight. The concept comes from research protections rather than from AI governance or model risk management, and its meaning is tied to the specific research and local context.

Formal definition

In the human subjects research context, minimal risk is commonly defined as the condition in which the probability and magnitude of physical, psychological, or informational harm or discomfort anticipated in the research are not greater than those ordinarily encountered in daily life or during the performance of routine physical or psychological examinations or tests. As commonly framed, its assessment is a calculus involving both the magnitude of a potential harmful outcome and the likelihood that the outcome will occur, and proper evaluation typically depends on knowledge of the local context in which the research is conducted. This determination frequently affects the level of ethical review applied and whether certain consent procedures are required. Note that this term as documented in the evidence originates in research ethics and is distinct from the term's use, if any, in AI risk classification schemes; the evidence provided does not address AI-specific 'minimal risk' categories, and that scope is not covered here.

Why it matters

Minimal risk functions as a gatekeeping threshold in human subjects research ethics: it helps determine how intensive an ethical review a study receives and whether certain procedures, such as documented written consent, are required. When research is judged to fall at or below this threshold, it may become eligible for expedited or otherwise less intensive oversight. Because the classification directly shapes the burden placed on researchers and the protections extended to participants, getting the determination right carries real consequences for both compliance and participant welfare.

A frequent source of error is treating minimal risk as a fixed, universal label attached to a type of activity rather than a context-dependent judgment. As the evidence indicates, procedures such as focus groups are ordinarily considered minimal risk, yet knowledge of the local context is critical to a proper assessment; the same procedure can carry different risk in different settings or populations. The determination is a calculus involving both the magnitude of a potential harmful outcome and the likelihood that it will occur, so collapsing it into a simple checklist can understate harms that are unlikely but severe, or overstate harms that are common but trivial.

For readers coming from AI governance or model risk management, the most important point is one of scope: this term originates in research protections and is not interchangeable with any 'minimal risk' category that may appear in AI risk classification schemes. The evidence here addresses only the research ethics meaning. Conflating the two could lead a practitioner to import assumptions from one domain into another where the definition, legal basis, and consequences differ.

Who it's relevant to

IRB members and human subjects research reviewers

Those who evaluate protocols use the minimal risk threshold to decide the appropriate level of review and consent procedures. The evidence underscores that this is a context-sensitive calculus of magnitude and likelihood rather than a fixed label, so reviewers benefit from attending to local context even for procedures ordinarily considered minimal risk.

Researchers designing human subjects studies

Investigators need to understand how a minimal risk determination affects oversight burden and consent requirements, including that much minimal risk research involves no procedures for which written consent would be required outside the research context. Understanding the threshold helps them anticipate the review pathway and document risks accurately.

Research compliance and ethics officers

Staff supporting research programs rely on a precise, context-aware understanding of the threshold to maintain consistent, defensible determinations. They should be alert to the risk of over-generalizing procedure-level assumptions without accounting for the specific research and local context.

AI governance and model risk practitioners encountering the term

Professionals from AI governance or model risk management should note that this term, as documented in the evidence, originates in research ethics and is distinct from any 'minimal risk' usage in AI risk classification schemes. The evidence provided does not address AI-specific minimal risk categories, so this concept should not be imported into AI risk frameworks without separate, domain-appropriate sourcing.

Inside Minimal Risk

Risk-tier classification

"Minimal risk" is commonly used as a risk category within tiered classification approaches to AI systems, most prominently associated with the EU AI Act's risk-based structure. In that framework it typically denotes the lowest tier, distinct from the unacceptable-risk, high-risk, and limited/transparency-risk categories. The exact category labels and their legal treatment are defined by the issuing instrument, so the term should not be assumed to carry the same meaning outside that framework.

Reduced or voluntary obligation set

Systems assessed as minimal risk are, in many risk-based frameworks, subject to few or no mandatory obligations beyond generally applicable law, with adherence to voluntary codes of conduct or good-practice measures often encouraged rather than required. The specific obligations that do or do not attach depend on the governing instrument and jurisdiction.

Classification as an assessed outcome, not an inherent property

Designation as minimal risk is the result of a risk assessment applied to a defined use case and context, rather than a fixed attribute of a technology. The same underlying model may fall into different tiers depending on its purpose, deployment context, and affected persons.

Relationship to inherent versus residual risk

A minimal-risk designation may reflect low inherent risk of the use case itself, and should be distinguished from residual risk that remains after controls are applied. The two are separate concepts, and a low tier assignment does not by itself indicate that controls have reduced a higher inherent risk.

Scope and boundary conditions

Minimal risk applies to a specific system and intended use as assessed; changes to purpose, data, user population, or deployment environment can move a system out of the category. The classification therefore has a defined scope and is not a permanent status.

Common questions

Answers to the questions practitioners most commonly ask about Minimal Risk.

Does a 'minimal risk' classification mean an AI system is exempt from all obligations?

No. As commonly framed in tiered risk approaches such as the EU AI Act, systems posing minimal (or low) risk are typically not subject to the mandatory requirements imposed on high-risk systems, but this is not the same as a blanket exemption. Certain baseline obligations, such as transparency duties in specific circumstances, may still apply depending on the system's function, and organizations often apply internal governance controls regardless of the external classification. Treat 'minimal risk' as a reduced-obligation tier under a particular framework rather than as an absence of any responsibility.

Is 'minimal risk' a universal category that means the same thing across every regulation and standard?

No. Risk-tiering language varies by instrument and jurisdiction, and a term like 'minimal risk' should be scoped to the specific framework using it. The categories in the EU AI Act are not interchangeable with the risk concepts in the NIST AI Risk Management Framework, ISO/IEC 42001, or model risk guidance such as SR 11-7, which use different vocabularies and pursue different objectives. Do not assume a classification under one regime carries over to another; each must be assessed on its own criteria.

How should an organization document why a system was placed in a minimal risk tier?

In many governance approaches, the classification rationale is recorded through a risk assessment that identifies the system's intended purpose, its context of use, the populations or decisions it affects, and the criteria in the applicable framework that led to the tier assignment. Documenting the assessment supports later review and challenge, particularly because a change in purpose or deployment context can move a system into a higher tier. The specific documentation format is not universally prescribed and often follows an organization's internal policy.

Who within an organization typically owns the decision to classify a system as minimal risk?

Classification decisions commonly involve the first line of defense (the business or development team that knows the system) proposing a tier, with second-line functions such as risk or compliance reviewing or challenging the assessment. The distribution of ownership depends on the organization's governance structure and the framework in use, and some organizations require independent sign-off for classifications that reduce the obligations applied. There is no single mandated ownership model across all frameworks.

How often should a minimal risk classification be revisited?

Because a risk tier reflects intended purpose and context of use, many governance programs reassess classification on a periodic basis and also upon material change, such as a new use case, expanded user population, or altered data. The appropriate cadence is typically defined in internal policy rather than fixed by a single external requirement, and a classification is best treated as a point-in-time judgment subject to revision rather than a permanent label.

What monitoring, if any, is appropriate for systems classified as minimal risk?

Even where a framework imposes few mandatory obligations on minimal risk systems, organizations often maintain lightweight monitoring to detect whether a system's use or impact has changed in ways that would warrant reclassification. Such measures reduce and help manage residual risk rather than eliminate it. The scope and intensity of monitoring is generally proportionate to the assessed risk and set by internal governance, not universally specified.

Common misconceptions

Minimal risk means the AI system is unregulated or free of any legal obligations.

A minimal-risk designation typically reduces framework-specific obligations, but generally applicable law (for example, data protection, consumer protection, and sector rules where relevant) can still apply. The extent of remaining obligations depends on the governing instrument and jurisdiction.

Minimal risk is a universal, interchangeable category across all AI frameworks.

The term is most closely associated with a specific tiered, risk-based approach and does not necessarily carry the same definition, thresholds, or consequences under other frameworks. Governance frameworks, voluntary standards, and model risk management guidance may use different terminology and categories, so the label should be scoped to its source.

Once classified as minimal risk, a system stays minimal risk.

Classification reflects an assessment of a defined use and context and can change if purpose, data, users, or deployment conditions change. A low tier is a point-in-time assessed outcome, not a permanent property of the system.

Best practices

Document the basis for a minimal-risk classification, including the assessed intended use, context, and affected persons, so the determination is traceable and reviewable.

Cite the specific framework or instrument under which 'minimal risk' is being applied, and avoid transferring the label or its consequences to other frameworks without re-assessment.

Re-evaluate the classification when the system's purpose, data, user population, or deployment environment changes, since such changes can move a system into a higher tier.

Distinguish inherent risk from residual risk in the assessment record so that a minimal-risk designation is not misread as evidence that controls have mitigated a higher underlying risk.

Confirm which generally applicable legal and sector obligations still apply, rather than assuming a minimal-risk label removes all requirements.

Consider adopting proportionate voluntary measures or codes of conduct where encouraged, treating them as risk-reducing rather than as eliminating residual risk.