ICRC Policy on Artificial Intelligence
Building a Responsible Humanitarian Approach: The ICRC's Policy on Artificial Intelligence
Purpose of the Policy
The ICRC's AI Policy provides guidance on the organization's practice and decision-making relating to the exploration, deployment, use, and management of AI and machine-learning systems. It sets out a value-based approach and provides a set of principles to promote a human-centred, responsible, and ethical use of AI in humanitarian action.
The guidance is aspirational and meant to support the ICRC's continuous efforts to use digital technologies responsibly, in line with its humanitarian mission, Fundamental Principles, values, and working procedures. Implementation will require investments in operational and governance capabilities, within the limits of the ICRC's capacities.
"By making this policy public, the ICRC wants to abide by our continuous commitment to be transparent and accountable for our actions." — ICRC AI Policy, November 2024
Definition & Scope
The ICRC defines Artificial Intelligence as: "any system, tool or technology that involves the use of computer systems to carry out tasks that would ordinarily require human cognition, planning or reasoning."
Machine Learning (ML) is defined as: "systems that use large amounts of data to develop their functioning and 'learn' from experience." For the purposes of this policy, the term "AI" refers to both AI and ML systems.
The policy complements the ICRC's binding Rules on Personal Data Protection and works in conjunction with the Handbook on Data Protection in Humanitarian Action. Complaints for non-compliance can be brought to the ICRC Data Protection Commission.
Why the ICRC Needs an AI Policy
The ICRC already uses AI tools to improve logistics, healthcare delivery, and data analysis. However, the rapid acceleration of AI innovations, simplified accessibility, and growing politicization of AI have created new risks and opportunities. The organization identified a critical gap: no common policy framework existed to guide AI-related workstreams across different departments.
The policy addresses both internal needs (coherent governance, consistent methodology) and external pressures (politicized AI debates, strategic competition between states, private sector risk-taking). It rejects both utopian and dystopian views of AI, instead adopting a pragmatic, humanitarian, and learning-oriented approach.
The Value-Based Approach
4.1 Humanity & Do No Harm
The ICRC's approach is driven by purely humanitarian objectives. AI is seen as a tool to alleviate suffering and improve operational efficiency — never as an end in itself. The organization systematically assesses potential negative secondary impacts on safety, dignity, agency, and fundamental rights.
🚫 What the ICRC Explicitly Prohibits
The ICRC will refrain from using, deploying, or providing AI systems that:
- Use subliminal, manipulative, or deceptive techniques to distort behaviour
- Exploit vulnerabilities related to age, disability, or socio-economic circumstances
- Use biometric categorization to infer sensitive personal attributes
- Engage in social scoring (evaluating individuals based on social behaviour or personal traits)
- Assess criminal risk based solely on profiling or personality traits
- Compile facial recognition databases via untargeted scraping
- Infer emotions except for medical or safety reasons
4.2 Impartiality
AI solutions must strengthen the ICRC's ability to respond to needs based on urgency and scale — without discrimination on race, nationality, gender, religion, or political opinion. The ICRC acknowledges that AI can replicate and amplify systemic biases (especially around gender and race) and commits to proactive mitigation.
4.3 Neutrality and Independence
AI is not neutral — it reflects the values of those who create it. The ICRC pays special attention to ensure that AI tools and providers cannot be instrumentalized to challenge the organization's neutrality. Where possible, the ICRC prefers open-source AI solutions to reduce dependency on politicized proprietary tools.
4.4 Enabling & Learning Approach
The ICRC commits to fostering ethical innovation, building AI literacy across all staff levels, and investing in training and upskilling. Access to AI tools may be restricted to specifically trained and authorized personnel when necessary.
The Four Guiding Principles
⚖️ 1. Proportionality, Precaution & Do No Harm
AI is only used when necessary and adequate to achieve a positive impact. Novelty or competitiveness alone are not justifications. A precautionary, science-based approach is applied, with risk levels classified as high, medium, or low. When benefits outweigh risks, solutions are tested in "sandbox" environments before rollout. Systematic auditing and feedback mechanisms are required.
🔒 2. Safety and Security
AI solutions are anchored in the highest possible levels of safety, security, and privacy. Risks of misuse, infiltration, and attacks are systematically considered. Data protection impact assessments are mandatory. Confidential or sensitive ICRC information cannot be used in non-approved external online AI systems.
🔍 3. Transparency & Explainability
The ICRC ensures the highest possible levels of transparency in AI use. Affected people and users are provided with intelligible information about when, why, and how AI systems impact them. This includes explanations of causal links between AI inputs and outputs.
📋 4. Responsibility and Accountability
Responsibility for AI-supported decisions always lies with the ICRC and its staff — never with the AI system. Affected people have the right to opt out of AI systems and access alternative solutions and complaint mechanisms. Environmental impact and carbon footprint must be considered throughout the AI lifecycle.
How the ICRC Compares to Other AI Governance Frameworks
Key Takeaways for Evaluators & M&E Professionals
📊 AI Risk Classification
The ICRC uses a differential approach distinguishing high, medium, and low-risk AI systems — a model that can be adapted for humanitarian M&E.
🔄 Mandatory Human Oversight
"Meaningful and effective human oversight mechanisms" are required for all AI systems impacting affected people — a core principle for ethical evaluation.
📝 Transparency & Explainability
Affected people have the right to understand how AI systems impact them — aligning with evaluation principles of stakeholder engagement.
🌍 Environmental Accountability
The ICRC is one of the few organizations explicitly requiring assessment of AI's carbon footprint — a growing consideration for sustainable M&E.
Key Sources that Informed the ICRC Policy
- UNESCO — Recommendations on the Ethics of Artificial Intelligence
- European Union — AI Act & Ethics Guidelines for Trustworthy AI
- The Alan Turing Institute — Understanding AI Ethics and Safety
- African Union — Continental AI Strategy
- ASEAN — Guide on AI Governance and Ethics
- Data Science and Ethics Group — Framework for the Ethical Use of Advanced Data Science Methods in the Humanitarian Sector
- Amnesty International & Access Now — The Toronto Declaration
- Chinese Academy of Information and Communication Technology — White paper on trustworthy artificial intelligence
Frequently Asked Questions
Does this policy apply to all ICRC staff?
Yes — it applies to all staff members and for all activities, internal services, operational responses, and partnerships involving AI technologies.
How does this relate to data protection?
The AI policy complements the ICRC's binding Rules on Personal Data Protection. Data protection impact assessments are mandatory before deploying any AI solution.
Does the ICRC use generative AI?
The policy covers generative AI, but with strict safeguards. Confidential or sensitive ICRC information cannot be used in non-approved external online AI systems (including public LLMs).
Can affected people opt out of AI systems?
Yes — provided it does not prevent them from benefiting from equivalent opportunities, the policy requires opt-out options and alternative solutions.
Why This Matters for the Humanitarian & Evaluation Community
The ICRC's AI policy is a pioneering framework that demonstrates how humanitarian principles can be operationalized for AI governance. For evaluators, it offers concrete guidance on:
✅ Risk-based classification of AI systems
✅ Mandatory human oversight and accountability
✅ Transparency and explainability requirements
✅ Prohibition of harmful AI practices (social scoring, manipulation, biometric surveillance)
As AI becomes increasingly prevalent in humanitarian action, this policy sets a standard for responsible, human-centred innovation.
Related Resources
