AI safety research has been concentrated in a handful of wealthy countries, leaving critical regional risks unaddressed. A new Global South AI Safety Hackathon, which concluded on June 21, 2026, brought together researchers, engineers, and policy professionals across Latin America, Africa, and Asia to tackle AI problems that matter most in their own regions. The event drew 1,507 sign-ups and challenged the assumption that AI safety is a one-size-fits-all problem.

Why Does AI Safety Look Different in the Global South?

The concentration of AI safety research in wealthy nations has created a blind spot. According to research cited in the hackathon materials, none of the top 100 institutions by AI publication index, in either universities or companies, are based in Africa or Latin America. This matters because AI risks manifest differently depending on local context, language, and infrastructure.

In the Global South, researchers face distinct challenges that Western-focused safety work often misses. Jailbreaks, which are attempts to bypass AI safety guardrails, are more common in low-resource languages where training data is scarce. Algorithmic bias trained on non-local data shows up in real-world deployments like healthcare and hiring systems, where the consequences can be severe. Africa, for example, faces risks from deepfake-driven electoral interference, data dependency on foreign infrastructure, and large-scale labor market disruptions that require locally-grounded solutions.

What Problems Did Hackathon Participants Focus On?

The hackathon was organized into three regional tracks, each with its own focus areas and winning teams. Participants competed within their region rather than globally, recognizing that solutions need to be tailored to local regulatory landscapes and institutional constraints.

• Latin America Track: Teams in São Paulo, Buenos Aires, Bogotá, and Mérida-Guadalajara worked on AI fairness for Portuguese and Spanish language models, safety evaluations for systems deployed in Latin American contexts, and regulatory analysis of emerging legislation. Brazil's AI bill includes a standalone human rights chapter that goes beyond the European Union's AI Act, and Chile became the first country in the world to constitutionally protect neuro-rights.
• Africa Track: Participants addressed misuse resilience, differential defense acceleration, and solutions to mitigate gradual disempowerment. Projects included policy recommendations grounded in African regulatory contexts and evaluations and benchmarks that transfer to African deployment scenarios.
• Asia Track: Teams in Bengaluru, New Delhi, and Vietnam tackled cross-border governance harmonization, safety evaluations for non-English language models, and region-specific risk assessments. Vietnam's AI law, which took effect in March 2026, represents the first binding AI legislation in Southeast Asia.

How to Participate in Future AI Safety Research Initiatives

The hackathon model offers a pathway for researchers and engineers in underrepresented regions to contribute to AI safety. Here are the key steps and opportunities:

• Form a Team: Participants can work individually or in groups, choosing a regional track and sub-track that aligns with their expertise and local context. Teams included founders, entrepreneurs, AI safety researchers, machine learning engineers, policy researchers, security researchers, and students.
• Scope a Problem: Using provided resources, teams research and define a specific AI safety problem relevant to their region. This might involve analyzing local regulatory gaps, identifying language-specific vulnerabilities, or assessing deployment risks in healthcare, hiring, or electoral systems.
• Build and Submit: Teams develop a tool, evaluation, policy analysis, or research contribution and submit a research report documenting their approach, results, and implications. The best teams are invited into the Apart Fellowship for continued research and mentorship.

The hackathon was supported by Schmidt Sciences and coordinated through regional hubs including BAISH in Buenos Aires, EA Brasil in São Paulo, AI Safety Colombia in Bogotá, AISMX in Mérida and Guadalajara, AI Safety South Africa in Cape Town, Electric Sheep in Bengaluru, Secure AI Futures Lab in New Delhi, and AnToàn.AI in Hanoi and Ho Chi Minh City.

Why This Shift Matters for AI Safety as a Field

The hackathon represents a fundamental reframing of AI safety work. Rather than exporting Western safety frameworks to the Global South, the event was designed to bring the Global South into AI safety research. Researchers in Latin America, Africa, and Asia possess contextual knowledge that the field desperately needs: understanding of regulatory landscapes, language gaps, and institutional constraints that determine whether safety research actually works in practice.

This approach acknowledges that AI safety cannot be solved by a small group of researchers in wealthy countries. As AI systems become increasingly global, the field must include voices and perspectives from the regions where these systems will be deployed and where their risks will be felt most acutely. The hackathon's structure, which invited winning teams into a fellowship program, creates a pipeline for sustained engagement and career development in AI safety research outside traditional hubs.

The event concluded with expert judges from AI safety organizations, universities, and policy institutions in each region reviewing submitted projects. Top teams received regional prizes and invitations to continue their work through the Apart Fellowship, signaling a long-term commitment to building AI safety capacity in underrepresented regions.