Learn AI /AI Governance and UN Terminology /Dual use
Dual use
- Intermediate
- AI Governance and UN Terminology
- Policy officers
- Human rights practitioners
Also called: dual-use technology.
Back to the milestone you came from
In five seconds
What it is
A tool, model or dataset can serve beneficial civilian purposes and also be adapted for harmful, coercive or military uses.
In plain words
One tool can be helpful in normal work and also be turned to harmful or military purposes, so rules must consider both sides.
An everyday example
Amina buys one sharp kitchen knife. In the morning, she uses it to cut vegetables for lunch. Later, a craft worker borrows it to trim thick packaging for a delivery. The same knife could also be misused to threaten someone, even though that was not its intended household purpose. People therefore store it carefully, decide who may use it, and pay attention to context.
A knife is a simple physical object, while AI systems are often made of software, data, models and surrounding infrastructure. AI capabilities can be copied at low cost, combined with other tools, scaled across many users and changed by new instructions. So the analogy helps show one tool with several uses, but it does not capture how quickly AI can spread or how much its behaviour depends on data, deployment setting and connected systems.
Remember thisDual use is not a fixed outcome. The same capability can enable legitimate civilian work or harmful, coercive or military applications, depending on tasking, users, data, access and controls.
This term is contested
The basic idea is widely used, but the exact boundary of what counts as dual use in AI can vary across policy, security, trade and human-rights discussions.
Understand it
Understand it
Why it matters
If you hear that an AI system is useful, that does not by itself tell you whether it is safe to share widely or deploy in every setting. Dual use matters because many policy choices are really about managing a capability that can help and harm at the same time. This affects export controls, procurement, access rules, transparency, testing, monitoring and accountability. For non-specialists, the key decision question is not only 'What can this do?' but also 'Who can use it, for what purpose, with which safeguards, and what could go wrong if it is repurposed?'
How one AI capability can lead to different uses
Show how the same capability can move into beneficial or harmful applications depending on access, context and safeguards.
01
Shared capability
Start with one capability, such as image analysis, route planning or code generation.
Draw one large rounded rectangle on the left labelled 'AI capability'. Add three small example tags beneath it: 'image analysis', 'route planning', 'code generation'.
02
Access and context
Who can use it, for what purpose, with what data and under what rules changes what the system enables in practice.
Draw a central rectangle labelled 'Access, data, purpose, safeguards'. Add four small icons or labels inside: user, data, goal, rules.
03
Beneficial path
With appropriate purpose and controls, the capability may support services such as translation, maintenance, research or humanitarian logistics.
From the central box, draw a green arrow to an upper-right box labelled 'Beneficial uses'. Add example bullets: 'translation', 'maintenance', 'logistics'.
04
Harmful path
With different access, weak controls or abusive intent, the same capability may support surveillance, cyber abuse or weapon-related functions.
From the central box, draw a red arrow to a lower-right box labelled 'Harmful or military uses'. Add example bullets: 'surveillance', 'cyber abuse', 'targeting support'.
05
Governance response
Risk management focuses on restricting dangerous uses while allowing legitimate ones where appropriate.
Place a final box spanning the right edge below both paths labelled 'Governance: assess risk, set limits, monitor use'. Connect both use boxes to it with arrows.
A left-to-right flow diagram with one shared capability entering a context-and-controls stage, then splitting into beneficial and harmful application paths. Read from left to right to see that the capability stays similar while the use changes with deployment conditions.
On hover or tap, each example tag reveals a short note showing how the same capability can appear in more than one sector; with keyboard focus, the same notes appear in a text panel.
Describe this without the diagram
This diagram explains dual use as a process. First, there is one shared AI capability, with examples such as image analysis, route planning and code generation. Next, that capability passes through a stage labelled access, data, purpose and safeguards. This middle stage shows that outcomes depend on who uses the capability, what data they apply, what goal they have and what rules limit them. From there, the diagram splits into two paths. One path leads to beneficial uses, with examples such as translation, maintenance and logistics. The other path leads to harmful or military uses, with examples such as surveillance, cyber abuse and targeting support. Both paths then connect to a final governance box that says assess risk, set limits and monitor use. The main message is that the same capability can support very different uses depending on context and controls.
With reduced motion: A static left-to-right diagram shows one AI capability, then a middle box for access, data, purpose and safeguards, then two branches: beneficial uses above and harmful or military uses below, both feeding into a governance box on the far right.
Dual use means the same technology can be used in more than one way, including a helpful use and a harmful, dangerous or military use. The idea matters in AI because many systems are general-purpose: they can support translation, research or logistics, but similar capabilities can also support surveillance, cyber operations or weapon systems.
What goes in is a capability: for example, image analysis, speech synthesis, route planning or code generation. What happens next depends on context, access, safeguards and who is using it. The same underlying capability may be combined with different data, hardware or instructions, leading to very different outcomes.
What comes out is not one fixed result but a governance problem. Policymakers, organisations and developers then ask questions such as: How risky is this use? Who should have access? What testing, oversight or restrictions are needed? In other words, dual use is about how a capability can travel across purposes.
This term is often confused with a technology being harmful by nature. That is not quite right. Dual use does not mean every use is equally likely or equally acceptable, and it does not mean a tool has no social value. It means benefits and risks can exist together, so governance has to look at both the capability and the context of use.
How “A powerful kitchen knife in different hands” maps onto the real thing
| In the story | In the AI system |
|---|---|
| one sharp kitchen knife | one capability that can serve several purposes |
| cutting vegetables and trimming packaging | beneficial civilian uses in different sectors |
| misusing the knife to threaten someone | harmful, coercive or military adaptation of the same capability |
| storing it carefully and controlling who may use it | access controls, safeguards and oversight |
A worked example
A developer offers a general-purpose speech recognition model that converts recorded speech into searchable text.
Starting from: Audio recordings, language settings and instructions for formatting the transcript.
- A public broadcaster uses the model to caption interviews and make its programmes more accessible.
- An abusive employer adapts the same model to transcribe private workplace conversations and search for criticism or organising activity.
- The core transcription capability remains substantially the same, while the tasking, users, data and consequences differ.
- Governance measures may therefore address access, bulk processing, retention, audit logs and monitoring for prohibited uses, rather than treating transcription as inherently benign or harmful.
Result: The model supports a legitimate accessibility function but can also enable coercive monitoring. Its dual-use risk becomes clear only when the capability is assessed together with who can use it, on which data and for what purpose.
In a United Nations workplace
Before rotating out, a human resources officer who serves as focal point for an AI-assisted document retrieval pilot in a documentation and library service adds a section to the handover note. The pilot extracts names, organisations and topics from public meeting records so desk officers can find relevant passages before close of business.
The same extraction and search components can support legitimate research or be retasked to map staff relationships and infer sensitive associations from personnel records. The handover therefore records approved data sources, access controls, logging, retention and escalation points. The risk follows from downstream tasking and context, not from a second, visibly different model.
Before you try thisPersonnel files, medical records, grievances and other confidential human resources material must not be uploaded to or indexed by a tool that has not been approved for those data. Doing so could expose personal data, enable profiling and breach confidentiality. Any expansion beyond public records requires formal review, restricted access and a documented purpose.
Where you may hear it
- An ICT governance board reviewing access to a general-purpose model
- Informals on draft General Assembly text concerning AI governance
- A procurement meeting testing a vendor's claims about misuse safeguards
- An ethics review before a research dataset is released
- A delegation briefing ahead of discussions on military applications of AI
Avoid confusion
Avoid confusion
Often confused with
Risk-based regulation
A way to set stricter rules for uses more likely to cause serious harm, while applying lighter duties to lower-risk uses.
Regulation
A regulation is a binding rule made by a competent authority under rule-making powers granted by law for those within its scope.
Lethal autonomous weapons systems
Weapons that can select and apply force to targets with reduced or no direct human control at the moment of attack, raising legal, ethical and security concerns.
Common misconceptions
Common misconception: If a technology has a beneficial use, calling it dual use unfairly treats it as suspicious.
Correction: The term does not say the technology should be banned or that its creators had harmful intent. It highlights that one capability may travel into different contexts, so governance should look at both public value and misuse risk.
Common misconception: Dual use only applies to military equipment.
Correction: The idea is strongly associated with military and security concerns, but it also applies to civilian technologies that can support coercion, crime, repression or other harm. In AI, general-purpose tools may be reused across many sectors.
Common misconception: A system is either civilian or military, so it cannot really be both.
Correction: A final product may be marketed for one sector, but the underlying capabilities can often be adapted, fine-tuned or combined with other systems for another purpose. That is why the same model or component may raise dual-use questions.
Quick check
Which statement best captures dual use in AI?
Dual use is about a capability serving different purposes in different contexts. The key issue is not only design intent or number of users, but whether the same capability can produce both beneficial and harmful applications and therefore needs appropriate safeguards.
Go deeper
Go deeper
The simple version
One tool can be helpful in normal work and also be turned to harmful or military purposes, so rules must consider both sides.
The technical version
In governance discussions, dual use refers to a capability, artefact or enabling component whose functional properties allow materially different downstream applications, including legitimate civilian uses and harmful, coercive or military uses. In AI, this often arises with general-purpose models, datasets, compute resources or software components that are not tied to one domain. The mechanism is not that the technology has one predetermined effect. Rather, deployment outcomes depend on tasking, model adaptation, connected systems, input data, user access, operational setting and institutional controls. Because the same capability can be repurposed across contexts, governance responses often focus on risk assessment, access management, use restrictions, monitoring, assurance and accountability instead of assuming a simple safe versus unsafe divide.
Reviewed external resources
Primary sourceGlobal Digital Compact (opens in a new tab)United Nations
This official UN page on the Global Digital Compact suits readers who want to see how the United Nations frames shared digital governance questions at a high level.
Intermediatelink checked 2026-07-31
Primary sourceGoverning AI for Humanity (opens in a new tab)UN Secretary-General's High-level Advisory Body on AI
This UN report suits readers who want policy-oriented discussion of AI governance challenges, including how benefits and risks may need to be managed together.
Intermediatelink checked 2026-07-31
EncyclopaediaDual-use technology (opens in a new tab)Wikipedia
This Wikipedia page gives a broad introductory overview of the concept across fields and suits readers who want a quick starting definition.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Dual use (opens in a new tab)YouTube search
This YouTube search results page suits readers who prefer video learning and want to compare several public explanations rather than rely on one source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Dual use (opens in a new tab)Reddit search
This Reddit search results page suits readers who want to see informal public discussion and differing interpretations, but it is not an authoritative source.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
Dual use is not a fixed outcome. The same capability can enable legitimate civilian work or harmful, coercive or military applications, depending on tasking, users, data, access and controls.