Event recap: Using AI Safely: Practical Strategies for CSOs and Nonprofits from a Global Majority Perspective

Through our support work at The Engine Room, we’ve been hearing growing questions from communities about AI, including its risks, opportunities, and how to use it responsibly. As part of our ongoing support for civil society organizations using the Cybersecurity Assessment Tool (CAT), we hosted a community call exploring practical and political approaches to AI safety from Global Majority perspectives. 

We were joined by three guest speakers: 

• Lupa, sociologist, systems administrator, and digital security consultant from TEDIC
• Ashi from Front Line Defenders 
• Chenai Chair, Director of Masakhane African Languages Hub

The speakers reflected on how AI  is being approached across Latin America and Africa, before moving into practical discussions about security, privacy, governance, language justice, and open-source alternatives.

Language AI and community-centered approaches in Africa

Masakhane African Languages Hub aims to build tools that support 1 billion Africans across 50 widely spoken African languages. Chenai, their director, emphasized significant talent and excitement around AI across the African continent, but uneven access to funding and resources persists. Many digital tools fail to recognize African accents, tones, and languages because developers largely design systems around colonial languages such as English, French, and Portuguese. Chenai stressed the importance of participatory approaches: communities should help create and evaluate datasets rather than having data extracted through large-scale scraping practices. Governments, researchers, and communities—not only private companies—should shape scalable and locally relevant AI systems.

AI, security and human rights defenders in Latin America

Lupa and Ashi, our two guests from Latin America, reflected on AI’s environmental and social impacts and stressed the importance of considering these before adopting new tools. A trend they have observed across the region is that human rights defenders tend to approach AI cautiously rather than rush to adopt it. While many people feel curious about AI, there is pressure to use it especially in resource and time constrained contexts. 

Attackers are increasingly using AI to strengthen phishing campaigns and other forms of digital harassment that target activists and marginalized communities. From a feminist and autonomy-focused perspective, speakers emphasized intentionally choosing how we use AI tools rather than letting them shape us. They urged taking ownership of tools to gain more autonomy by understanding why you are using a specific tool, what for and how the tool is going to give you the desired outcome.

Safety tips and recommendations

Our guest speakers shared a set of practical, grounded tips for engaging with AI in ways that prioritize safety, autonomy, and context. These tips can guide you as you navigate increasingly complex digital environments. 

Protect sensitive data 

• Practice data minimization before sharing information with AI tools. Share only what is strictly necessary to fulfill the specific task.  
• Avoid uploading confidential organizational or community data to AI tools
• Review and adjust AI tool privacy and security settings, including disabling data training where possible
• Check data retention policies to understand how long AI tools store your data
• Consider using mainstream tools without accounts to reduce data exposure

Verify tools and outputs 

• Verify AI-generated information before relying on it. AI can hallucinate or spread misinformation
• Evaluate whether tools genuinely meet your needs or simply collect data
• Consider whether AI is actually necessary for the task

Prioritize privacy and autonomy 

• Use privacy-focused tools such as Duck.ai or Lumo where possible. 
• Be mindful of tracking through cookies and location data. Use tools such as VPNs and Tor to minimize tracking. You can also regularly clear cookies to maintain privacy
• Explore open-source or self-hosted alternatives for greater control 
• Be mindful of dependence on infrastructure controlled by large tech companies

Use AI tools strategically

• AI should support human work, not replace it
• Define a clear, meaningful use case before adopting any tool
• Avoid over-reliance. Many problems can still be solved effectively without AI
• Balance automation with human judgment and expertise
• Regularly assess whether a tool is truly useful or simply adding complexity

Build community-centred AI practices

• Include community feedback, recommendations, autonomy and consent in AI dataset design and evaluation
• Consider disability, gender, language, and age inclusion
• Address extractive data practices and labor inequalities in AI development. 

Localize and contextualize AI systems

• Use locally relevant benchmarks and evaluation methods. Many AI systems are built using Global North assumptions, which may not translate effectively in other contexts.
• Build datasets and models that reflect local languages and realities
• Question whether existing systems are appropriate for your context

Questions to ask before adopting an AI tool

• Does the tool align with your values and political context?
• What is the tool’s reputation, and with whom and on what kinds of partnerships or agreements has the company recently entered into?
• Have there been any recent policy changes?
• Where are its servers located?
• Is it based on small language models or large language models?
• Is AI required for this use case, or can this be done using a different type of tool, or simply by talking to other people?
• Is it transparent about data usage and privacy? How long does the tool store and retain our data?
• Does the tool use a Big Tech company’s infrastructure?

We look forward to continuing these conversations through future community calls and shared learning spaces for digital security practitioners and organizations seeking support. If you missed the call and would like to continue the conversation, reach out to our support team at cybercat@theengineroom.org 

You can also book a CATio, a space dedicated to supporting at-risk organizations globally with the Cybersecurity Assessment Tool (CAT).