Strengthening AI Incident Monitoring and Reporting in Africa for Global AI Safety

The past few years have been marked by a series of AI safety and security incidents, ranging from multiple AI agents orchestrating severe cyberattacks to actors misusing AI models to plan mass shootings, bombings and terrorist attacks that have resulted in deaths and injuries. These incidents are only a fraction of the rising tally of AI-related harms. The number of AI incidents catalogued by AI incident reporting databases have been rising steadily over the years. The aim of these databases is to track both realised harms (AI incidents) as well as near-harms and potential harms (AI issues/hazards), and by doing so, reduce the likelihood of occurrence and recurrence of AI harms. 

Though AI incident databases now capture thousands of AI incidents, their coverage remains geographically uneven. Between February 2020 and July 2026, for example, AI incidents from African countries averaged only ~ 1.3% of the total AI incidents reported in the OECD AI Incidents and Hazards Monitor (AIM), and ~ 4.2% of the AI incident records in the AI Incident Database (AIID), with most of these AI incidents being information-related harms such as deepfakes, scams and fraud. This underrepresentation is not a failure of the databases. AIID, for example, strives to make their reporting as global and comprehensive as possible by scanning regional sources, including African news sources and fact-checking services. The gap is structural: because the databases rely on publicly reported incidents, harms that never go public go unreported. Many harms arising in the African continent are likely to be part of these underreported incidents, as monitoring of AI harms in Africa is still severely underdeveloped. This underdevelopment also means that harms that are hard to discover, such as harms that are diffuse, systemic or occur in less visible institutional settings are also likely to be underreported in Africa, plausibly explaining the dominance of information-related harms, which are easier to discover, in the AIID.

Yet, the continent is seeing increased AI adoption by governments, corporations and individuals, which might result in more AI incidents. Comprehensive monitoring and reporting of these incidents is important not just for assuring the safety of Africans but also ensuring global AI safety. 

AI incident information matters for African safety

Strengthening AI incident monitoring and reporting in Africa is essential for ensuring the safety of a continent that currently hosts more than 1.5 billion people. As AI capabilities continue to scale rapidly and adoption becomes more widespread across Africa and globally, this large population is exposed to extreme risks from AI, including extreme manipulation, increased armed conflict, and more enduring and repressive authoritarianism, among many others. Indeed, there are already reports of terrorist groups in Africa using AI to plan attacks, and to design and troubleshoot weapons. These harms may arise not just from products deployed in Africa but also from products deployed or used elsewhere but cause harm in Africa. As AI cyber capabilities continue to increase, for instance, global threat actors outside Africa might use AI to launch cyberattacks against targets in Africa. 

Comprehensive AI incident monitoring and reporting is needed to capture these harms and provide critical safety-learning information. Developers can learn from this information and adjust their design, development and deployment processes to provide safer AI products for the African market and prevent future incidents. Importantly, African governments can also use this information to put in place policies and measures to guard their populations against future harms. Strengthening AI incident reporting is thus key to securing the safety of Africa. 

AI incident information from Africa matters for global safety

AI incident information from Africa is useful not just for promoting safety in Africa, but also global AI safety. This information can serve as an early warning sign for other regions. Many AI models deployed in Africa are deployed globally, hence failures from these models in the African context might occur in the other regions. Information on incidents in Africa might highlight the underlying model capabilities, and the vulnerabilities in safeguards that led or contributed to the incidents and which could lead to the same failures in other regions. This can thus help other regions in incident prevention and preparedness. Furthermore, as AI adoption on the continent continues to grow and AI capabilities improve significantly, so does the risk that threat actors currently causing harm in Africa may turn to other parts of the world. Hence, monitoring the incidents that occur in Africa can help prevent escalation of the harms to other regions. Two concrete examples illustrate this clearly:

Terrorist use of frontier AI in Africa

In July 2026, a report uncovered the extensive use of frontier AI models by the Boko Haram terrorist group (which operates in the Lake Chad Basin area in West and Central Africa) in attack planning, weapons design and troubleshooting, and improving operational security. The report highlighted how the terrorist group has managed to bypass model safeguards in order to achieve these objectives. Notably, the implicated frontier AI models ⸺ ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek ⸺  are widely deployed and accessible across the world, presumably including to other terrorist groups. 

Such evidence of how terrorists in Africa are using AI is important for ensuring the safety of the affected African communities and the rest of the world. It is useful in ‘informing threat models, identifying vulnerabilities in model safeguards, and establishing a baseline for tracking how terrorist use of frontier AI evolves over time’. These are insights whose utility extends beyond the African context. Model vulnerabilities exploited by terrorists in Africa, for example, are likely to be exploited by terrorists in other regions. This is very likely as major terrorist groups like Boko Haram are critically integrated into, or associated with, global terrorist networks such as al-Qaeda and the Islamic State which have a history of systematic knowledge and resource transfer that now extends to how to exploit frontier AI capabilities in terrorist activities. Similarly, Islamic State-Somalia (ISIS-S), which has used Somalia ‘as a testing and refinement environment’ for using commercial unmanned aerial systems for terrorist activities and shared this knowledge with other ISIS affiliates, might share its knowledge on exploiting frontier AI capabilities with other ISIS affiliates. Understanding how terrorists in Africa use AI can thus highlight how other terrorist groups within these networks are using, or are likely to use, AI.

Information on how terrorist groups in Africa are using AI could also be important for other countries if these terrorists have global ambitions, as Boko Haram does, for example, and especially if models provide meaningful uplift that enables execution of these global ambitions. This is especially if AI significantly uplifts the terrorists’ ability to develop and deploy chemical, biological, radiological and nuclear (CBRN) weapons, which some members of Boko Haram have expressed willingness to use, or at least are not categorically opposed to. The harms that are already manifesting in Africa thus illustrate the risk that the rest of the world faces if threat actors that have traditionally been limited to Africa gain sufficient uplift from AI to expand their targets beyond Africa. 

Low-resource language jailbreaks

Research has found that models tend to be more unsafe in low-resource languages, including through providing users with advice on harmful objectives such as building bombs and financial fraud. There are many low-resource languages spoken in Africa which safety mechanisms do not generalize to and can therefore be used to jailbreak AI models to cause harm. Harms arising from such misuse might be more prominent within Africa, which hosts most speakers of these languages. However, there are many speakers of these languages residing outside of Africa, who might also misuse AI through language jailbreaks. As such, although these harms might be concentrated in Africa, they might not be confined to it, and monitoring the harms in Africa can help prevent them from occurring elsewhere. 

Conclusion

Given the safety benefits of AI incident information from Africa, there is need to invest in and support efforts to strengthen AI incident monitoring and reporting in Africa. This includes supporting local actors best placed to detect harms as they occur, such as researchers, journalists, civil society organisations and African governments. For the AI incident information arising from this to support global AI safety, there is also need to support cross-border AI incident information sharing, through mechanisms such as the International Network of Advanced AI Measurement, Evaluation, and Science, as well as through bilateral agreements with national AISIs and the EU AI Office. 

Call to Action

For local researchers, journalists and civil society: Monitor and document AI safety incidents arising on the continent. You can also submit incidents you come across to the AI Incident Database.

For African governments: Invest in national-level AI incident monitoring to detect harms, and collaborate with other governments and actors to share relevant incident information.

For funders and international community: support and/or collaborate with local actors best placed to detect AI harm, and back efforts to establish cross-border AI incident information sharing. 

- - - 

Marie Iradukunda is a Research Associate at the ILINA Program, where her work focuses on the governance of frontier AI in Global South countries, with a particular focus on monitoring and reporting AI harms. She holds a Bachelor of Laws degree from Strathmore University and a Master of Laws degree from Harvard Law School.

Acknowledgements

Thank you to Daniel Atherton and Sean McGregor for their insightful comments on this post.