Mentees will analyze concrete AI-induced crisis scenarios using a structured failure-mode decomposition template and extract the response capacities that governments, labs, and infrastructure operators would need to mitigate harm when prevention fails. The goal is to identify which capacities recur across diverse scenarios and where current institutions are structurally unprepared.
About the project
Frontier-AI governance focuses on prevention, but failures can still occur, and we lack a clear picture of where systems and institutions are structurally vulnerable once harm begins to unfold. This project targets that gap.
Mentees will choose a near- to mid-term AI-induced crisis scenario (e.g., model escape, AI-enabled cyber cascades, scaled criminal misuse, mass-persuasion shocks, concentrated psychological harm). Using a vulnerability-analysis template, they will identify the key stressors the scenario imposes, the points where institutions, coordination, or information flows fail under those stressors, and the concrete capacities that would be required to absorb or mitigate each vulnerability (state capacity, monitoring, surge response, legal authorities, cross-sector communication, etc.).
Each mentee produces a scenario-specific vulnerability map and a short list of required response capacities. The project then compares findings across scenarios to identify capacities that recur across different crisis types. These repeated needs highlight structural gaps in current institutions and indicate where preparedness efforts would yield the highest payoff.
The work is descriptive rather than predictive: no technical modeling or policy design. Analysis draws on historical analogues from domains such as nuclear safety, cybersecurity incidents, pandemics, and financial crises to ground judgments about institutional weaknesses and response dynamics.
Theory of change
AI safety discussions are heavily weighted toward preventive controls on developers, but failures can still occur under deep uncertainty. If prevention fails in a non-catastrophic but high-impact way, the world currently has little shared understanding of what effective crisis response would require. By systematically mapping AI-induced crisis scenarios and extracting shared response capacities, this project identifies the concrete institutional capabilities that matter most once harm is unfolding. This work fills a gap between technical safety research and high-level policy concepts by generating actionable knowledge about cross-scenario robustness and the state capacities needed to prevent local failures from escalating into systemic ones.
Your role
Mentees will contribute to a larger ongoing research effort on AI-related crisis response and coordination capacity. Each mentee will independently analyze one or more AI-induced crisis scenarios using a structured failure-mode decomposition template. The work is parallelizable, meaning mentees choose a scenario(s), conduct the analysis, identify required response capacities, and document their findings. These outputs will be integrated into a shared synthesis mapping cross-scenario response needs. Contributions are time-bounded, and individual work will be clearly acknowledged in the final output of the broader project. The project is embedded within a wider research context with additional collaborators, though formal commitments lie only with the listed mentor.
Prerequisites
- Familiarity with AI-related risks (technical, governance, or societal).
- Strong analytical writing skills.
- Ability to reason clearly under uncertainty and maintain rigorous epistemic standards.
- No specific disciplinary background required; applicants from policy, social science, technical fields, philosophy, security studies, or adjacent areas are welcome.
Time commitment
5-8h per week.
Location preference
No geographical preference. Meetings will be scheduled within CET working hours.
Application question(s)
-
Choose one AI-induced crisis scenario. Do a short literature scan and identify two papers you consider most useful for understanding the structural stresses or failure conditions relevant to this scenario. For each paper, explain in 2–3 sentences why it is relevant for identifying institutional or system vulnerabilities. (max 250 words, excluding citations)
-
Suppose you could strengthen only one vulnerability you identified. Which one would you prioritize, and why is it more pivotal than at least one other plausible option? (max 150 words)
-
Identify three core vulnerabilities in this scenario. For each, briefly state: (a) what stressor exposes it, and (b) what concrete capacity would be required to mitigate or absorb that vulnerability. Focus on clarity and mechanism. (max 200 words)
About the mentor
Isaak Mengesha studies how societies adapt to fast technological change. Trained in computational economics and complex-systems modeling, he has worked on technological transitions, growth constraints, and structural vulnerabilities in development. He now brings this perspective to AI governance, focusing on how to reason and plan under uncertainty.