Skip to main content
Bureau of AI
Category: Trustworthy AI Principles

Harm Prevention

Also known as: Harm Reduction
Simply put

The evidence provided defines only 'harm reduction,' a public-health approach that seeks to lessen the negative health, social, and legal consequences associated with drug use rather than requiring people to stop using. It is described as a set of practical strategies and public health interventions that 'meet people where they are,' accepting that abstinence may not be feasible or immediately achievable for everyone. The evidence packet does not establish 'Harm Prevention' as a defined term, nor does it connect either term to AI governance or model risk management.

Formal definition

As documented in the supplied sources, harm reduction is a public health strategy consisting of policies, programmes, and practices intended to minimise the negative health, social, and legal impacts associated with drug use, developed initially for adults with substance use problems for whom abstinence was not feasible. The evidence does not support a distinct, established definition of 'Harm Prevention' as a term of art, and no cited material treats it as a distinct concept. Critically, none of the provided sources address artificial intelligence, AI governance, trustworthy AI principles, or model risk management; any application of this public-health concept to AI contexts would be an analogy unsupported by the evidence packet. This entry should therefore be understood as scoped to public-health drug policy only, and its terminology (title versus the term actually evidenced, 'harm reduction') and any AI relevance remain unverified pending sources specific to that domain.

Why it matters

The evidence packet supports only the public-health concept of "harm reduction" as it applies to drug use, and does not establish "Harm Prevention" as a defined term or connect either term to AI governance or model risk management. For readers of a publication focused on AI governance, this distinction matters: the term as titled cannot be treated as an established concept within trustworthy AI principles, model risk management, or any AI regulatory framework on the basis of the sources provided. Applying the public-health notion of harm reduction to AI contexts would be an analogy, not a documented practice, and the evidence does not support drawing that analogy.

Who it's relevant to

Public health and drug policy professionals
Based on the evidence, harm reduction is relevant to those working in public health, substance use treatment, and drug policy, where it functions as a strategy to reduce the negative health, social, and legal impacts of drug use. This is the only domain the cited sources address.
AI governance and model risk professionals (with caution)
The evidence packet does not connect harm reduction or "Harm Prevention" to AI governance, trustworthy AI principles, or model risk management, and cites no AI-specific instruments such as the EU AI Act or the NIST AI Risk Management Framework. Professionals in these areas should not treat this entry as an established AI governance term; any relevance to AI would require sources specific to that domain, which are not present here.
Editors and terminology reviewers
Because the title ("Harm Prevention") differs from the term actually evidenced ("harm reduction"), and because the assigned category does not match the public-health scope of the sources, this entry should be flagged for terminology and categorisation review before publication.

Inside Harm Prevention

Scope caution
The available context did not establish 'Harm Prevention' as a defined term of art within AI governance or model risk management. The cited material addressed a public-health concept (harm reduction in drug policy) rather than AI-specific usage, so no verified components of an AI governance term can be enumerated here.
Distinction from harm reduction
Harm prevention, in general usage, refers to measures aimed at avoiding an adverse outcome before it occurs, whereas harm reduction refers to measures aimed at lessening the severity of adverse effects that are expected to occur regardless. The context conflated these, and this distinction should not be assumed to carry a settled AI-specific meaning.
Relationship to risk management (unverified)
In many AI governance and model risk management settings, controls are commonly framed as reducing or managing risk rather than eliminating it. Whether 'harm prevention' is a recognized label for such controls in any specific framework could not be confirmed from the provided material and should be treated as unsupported until sourced.

Common questions

Answers to the questions practitioners most commonly ask about Harm Prevention.

Is "harm prevention" an established, defined term in AI governance frameworks?
Not in the way the phrase might suggest. The available sources for this concept come from public-health contexts (notably harm reduction in drug policy) rather than from AI-specific regulatory instruments. There is no evidence presented here that "harm prevention" is a settled, standardized term within AI governance or model risk management frameworks. Practitioners should treat it as a general-purpose concept rather than a codified governance requirement, and should verify how any specific framework or organization defines the phrase before relying on it.
Can the public-health definition of harm reduction be treated as a model for AI governance?
This should be approached with caution. The concept as documented derives from public-health and drug-policy contexts, and extending it to AI governance—even by analogy—goes beyond what those sources support. AI governance frameworks address organizational structures, accountability, and oversight of AI systems, which is a distinct domain. Borrowing terminology across fields risks conflating unrelated guidance, so any such analogy should be flagged as informal rather than presented as an established governance principle.
What should an organization confirm before adopting "harm prevention" language in its AI policies?
An organization should first confirm whether the term maps to a defined obligation in any applicable framework, since the phrase is not, on the available evidence, a standardized AI governance term. Where the phrase is used internally, the policy should specify its intended meaning, the risks it is meant to address, and how it relates—if at all—to recognized instruments the organization is subject to. Documenting this scope helps avoid ambiguity during audit or review.
How does one avoid conflating public-health guidance with AI regulatory requirements when using this term?
Keep the sources of each concept distinct. Public-health guidance and AI regulatory instruments are issued by different bodies for different purposes, and none of the public-health material cited here establishes AI-specific requirements. In practice, this means citing AI-relevant authorities only where they genuinely apply, and not implying that a public-health framing carries regulatory weight in an AI context. Where the connection is only conceptual, label it as such.
Where does responsibility for concepts like harm prevention typically sit within an organization's lines of defense?
As commonly framed, operational ownership of a risk-related objective typically sits with the first line of defense (business and model owners), with the second line (independent risk and compliance functions) setting policy and challenging implementation, and the third line (internal audit) providing independent assurance. Because "harm prevention" is not a codified term in this context, an organization would need to define which line owns the objective, rather than assuming a standard allocation exists.
How can a compliance team document its handling of a term that lacks a settled definition?
The team can record a working definition, note that the term is not established in cited AI governance sources, and identify any contested or evolving treatment. Documentation should distinguish what is a recognized requirement from what is an internal or analogical usage, and should state what is out of scope. This transparency supports later review and reduces the risk that informal terminology is later read as an implied regulatory obligation.

Common misconceptions

'Harm prevention' is an established, well-defined term in AI governance frameworks.
No evidence in the provided context supports treating 'harm prevention' as a defined AI governance term. The cited sources concerned public-health drug policy, not AI regulation or model risk management, so any AI-specific definition here would be unsupported.
Governance or risk controls can prevent harm entirely.
As commonly framed in AI governance and model risk management, controls typically reduce or manage the likelihood and severity of harm; they do not eliminate risk. Describing measures as fully 'preventing' harm overstates what such controls can achieve.
Public-health harm-reduction concepts translate directly into AI governance requirements.
Public-health guidance on drug policy operates in a different domain and does not, on its own, establish obligations, definitions, or best practices for AI systems. Instruments such as the EU AI Act (issued by EU institutions) or the NIST AI Risk Management Framework (a voluntary U.S. framework) are distinct in scope, authority, and jurisdiction and were not cited in support of this term.

Best practices

Before adopting 'harm prevention' as a defined concept, confirm whether it appears as a term of art in the specific framework you are applying (for example, a named regulatory instrument, a voluntary standard, or supervisory guidance), and cite that source directly rather than relying on analogy from other domains.
Do not import public-health concepts such as harm reduction into AI governance without a sourced basis; keep domain-specific meanings separate and flag where a term's definition is contested or unverified.
Frame any controls described under this concept as measures that reduce or manage risk, using qualified language rather than implying that harm is eliminated.
Scope each cited instrument to its issuing body and legal status (binding law, supervisory guidance, or voluntary standard) and avoid implying that frameworks like the EU AI Act, NIST AI RMF, ISO/IEC 42001, or SR 11-7 / OCC 2011-12 are interchangeable or universally applicable.
When evidence is limited to an unrelated domain, state the limitation explicitly and decline to fabricate AI-specific components, definitions, or regulatory references.
Distinguish prevention (avoiding an outcome before it occurs) from reduction (lessening severity of an expected outcome) in any documentation, so reviewers and auditors are not misled by loose terminology.