NetHope’s Humanitarian AI Code of Conduct
NetHope's Humanitarian AI Code of Conduct: A Sector-Wide Ethical Framework
Why a Humanitarian AI Code of Conduct?
The rise of artificial intelligence technologies, particularly those powered by machine learning, large language models, and generative AI, presents both significant opportunities and serious risks for humanitarian organisations. While AI can help organisations make better use of data, automate processes, create efficiencies, and improve operations and analysis, the risks have been well documented.
The core challenge:
Without sector-wide agreement on how to engage ethically and fairly with AI technology, humanitarian organisations risk engaging in a "race to the bottom" — constantly chasing new tools while engaging in potentially dangerous activities that put the people they are meant to support at risk of harm.
NetHope's Humanitarian AI Code of Conduct addresses this gap. Developed by NetHope's Member-led AI Working Group and authored by Elizabeth Shaughnessy, NetHope's Director of Digital Programming, the code provides a collective framework to help organisations navigate the ethical complexities of AI. It is designed to support internal policy development, align with rapidly evolving legislation, and enable robust governance, operations, programming, and advocacy across the sector.
Understanding the Risks: Why Guardrails Are Essential
The code is explicit about the harms that AI technologies can cause, particularly for vulnerable populations in humanitarian contexts:
The Thirteen Principles of the Humanitarian AI Code of Conduct
The code is organised into three sections: foundational commitments for all AI use, specific prohibitions for high-risk concerns, and collective actions for sector-wide collaboration.
Foundational Commitments (Principles 1-3)
Principle 1: Do No Harm
Our use of AI technology and related data practices must do no harm and align with existing humanitarian principles and frameworks.
Principle 2: Net Positive Impact
AI use must have a net positive impact on our organisations' missions and for the individuals and communities we serve, and not exacerbate the problems we are working to solve — including inequality, conflict and fragility, and climate change.
Principle 3: Fair, Inclusive, Accessible and Feminist
AI use must be fair, inclusive, accessible, and feminist; mitigate bias; and ensure transparency and explainability. An intersectional feminist approach to AI requires an analysis of power and impacts on social justice.
High-Risk Safeguards (Principles 4-7)
Principle 4: No Synthetic Media of Vulnerable Groups
Organisations agree not to use AI to generate photo-realistic images or videos (or other related content such as voice or audio samples) of vulnerable groups, including children and program participants, for the purposes of publication, including campaigning and fundraising.
Principle 5: Human Review of Published Content
A human will review and, where necessary, contextualize and attribute content generated by or with the help of AI for the purposes of publication.
Principle 6: Human Review of AI-Made Decisions
A human will review decisions made or facilitated by AI, in particular where there is risk of real or perceived harm to an individual or community. Those affected have a right to human review of AI-made or facilitated decision-making.
Principle 7: Prioritise Safeguarding and Child Protection
Organisations will prioritise safeguarding and child protection and establish any additional guardrails needed where AI is used in these contexts. For example, agreeing not to use facial recognition technology for those under 18 years of age.
Collective Commitments (Principles 8-13)
Principle 8: Collaborate and Exchange Knowledge
Organisations will collaborate and exchange knowledge and best practices, and collectively support digital literacy and skills in AI across the sector.
Principle 9: Align on Due Diligence and Procurement
Organisations will align on the approach to due diligence, assessment, and procurement of AI vendors and tools, as well as limits on the commercial use of data.
Principle 10: Uphold Highest Data Protection Standards
Organisations will align across different regulatory environments and uphold the highest available standards and practice of data protection, privacy, and security used by model.
Principle 11: Address Informed Consent in AI Contexts
Organisations will understand and address how informed consent manifests in AI contexts and whether it can be meaningfully used as a basis for collecting personal data.
Principle 12: Ensure Governance, Accountability and Redress
Organisations will ensure effective governance, accountability, and pathways for redress for those harmed by AI systems.
Principle 13: Engage with Emerging Risks Collectively
Organisations will collectively engage with questions and address new risks as the AI landscape evolves, including on the commodification of data, the ability to opt out, and the right to be forgotten.
Key Takeaways for Evaluators and M&E Professionals
Human review is non-negotiable
Principles 5 and 6 establish that humans must review both AI-generated content intended for publication and AI-made decisions that could cause harm. This has direct implications for how M&E functions assess AI use in programmatic decision-making.
Safeguarding requires additional AI guardrails
Principle 7 explicitly calls for prioritising safeguarding and child protection, including prohibitions on facial recognition for minors. Evaluators should assess whether organisations have implemented these additional guardrails.
Net positive impact must be measured
Principle 2 requires that AI use have a net positive impact and not exacerbate problems like inequality or climate change. This creates a clear M&E requirement: organisations must be able to measure and demonstrate net positive impact.
Transparency and explainability are required
Principle 3 demands transparency and explainability. Evaluators should look for documentation of how AI systems make decisions and whether those decisions can be explained to affected communities.
A unique moment in the timeline of AI development
The code's introduction makes a powerful argument: the humanitarian sector has a unique opportunity in the timeline of AI development to thoughtfully and collaboratively work together to address and mitigate risks, demystify AI technology, and set responsible standards before the use of such technology becomes ubiquitous. By adopting this code, organisations can support internal policy development, align with rapidly evolving legislation, and prepare themselves for robust governance, operations, programming, and advocacy in relation to AI technologies.
Supporting Resources from NetHope
NetHope has produced a number of resources to help organisations mitigate AI risks and implement the principles of this code:
- AI Ethics Train-the-Trainer program — building internal capacity for ethical AI implementation
- Gender Equitable AI Toolkit — addressing bias and promoting feminist approaches to AI
- NetHope AI Lighthouse — a guide to responsible, purpose-driven AI specifically for nonprofits
- Data Governance Toolkit for Nonprofits — establishing data foundations critical for AI projects
Frequently Asked Questions
Who developed this code and who should use it?
The code was developed by NetHope's Member-led AI Working Group, authored by Elizabeth Shaughnessy, NetHope's Director of Digital Programming, and supported by Daniela Weber, Director of NetHope's Center for the Digital Nonprofit. It is designed for humanitarian, development, and conservation organisations of all sizes, as well as their partners and vendors.
What does "feminist AI" mean in Principle 3?
An intersectional feminist approach to AI requires an analysis of power and impacts on social justice. The code references the Feminist Principles of the Internet and Data Feminism as primers. This means actively considering how AI systems may perpetuate gender-based discrimination and ensuring that AI development and deployment include diverse perspectives, particularly from those most marginalised.
Does the code prohibit all use of synthetic media?
No. Principle 4 specifically prohibits generating photo-realistic images or videos of vulnerable groups (including children and program participants) for publication, campaigning, or fundraising. It does not prohibit other uses of synthetic media, but any such use would still need to comply with other principles including human review (Principle 5) and do no harm (Principle 1).
Is this code legally binding?
The code is a voluntary, sector-wide commitment rather than a legally binding instrument. However, it is designed to align with rapidly evolving legislation and governance efforts. Organisations are encouraged to adopt the code's principles into their own internal policies, procurement processes, and partnership agreements.
A collective commitment to responsible AI
The NetHope Humanitarian AI Code of Conduct represents a significant step forward for the sector. By agreeing to these thirteen principles, organisations commit to a shared vision of AI that serves humanity rather than harming it — one that is fair, transparent, accountable, and centred on the dignity and rights of the people they serve.
As the code states: "Together, we can prepare ourselves and support each other to have robust governance, operations, programming, and advocacy in relation to AI technologies."
For evaluators and M&E professionals, the code provides a clear benchmark against which to assess organisational AI readiness, governance structures, and ethical compliance.
Related Resources
