MIT AI Risk Initiative: Prioritization of Risks From Artificial Intelligence
Following our previous exchanges and cooperation, we contributed to the MIT AI Risk Initiative's Prioritization of Risks from Artificial Intelligence study as co-authors, focusing on our work related to physical and embodied AI systems, advanced sensing and perception. The study involves 272 authors from 37 countries and covers the prioritization of 24 AI risk domains across probability, severity, sector vulnerability, actor vulnerability, and responsibility.
Related
MIT AI Risk Initiative
The MIT AI Risk Initiative, operating within MIT FutureTech at the Massachusetts Institute of Technology, advances structured, evidence-based analysis of risks arising from artificial intelligence systems. Situated at the intersection of technical AI research, governance scholarship, and policy, the Initiative engages interdisciplinary and international expert communities to systematically assess AI risks across domains and sectors. Its work builds on established risk governance frameworks, including the NIST AI Risk Management Framework and ISO 31000 standard for risk management, and engages with the broader landscape documented in the International AI Safety Report.
Prioritization of Risks from Artificial Intelligence
Conducted as a three-round Delphi study across 272 international experts from academia, industry, government, and civil society spanning 37 countries, the study provides a structured assessment of 24 AI risk domains, synthesized from 74 existing frameworks, across two scenarios: business-as-usual and pragmatic mitigations. Experts rated each risk domain on harm probability and severity, sector and actor vulnerability, and actor responsibility, offering one of the most comprehensive multi-stakeholder expert elicitations on AI risk prioritization to date.
Under a business-as-usual scenario, experts assigned at least a 10% probability of catastrophic outcomes, defined as more than one million human deaths, more than USD 100 billion in financial loss, or equivalent civilizational-scale intangible harm, to 18 of 24 risk domains over the period 2025–2030. The five highest-severity risks were identified as dangerous capabilities, competitive dynamics, weapons and cyberattacks, power centralization, and false information. Even under a pragmatic mitigations scenario, five risks, including dangerous capabilities, weapons and cyberattacks, environmental harm, inequality and unemployment, and power centralization, retained catastrophic harm probabilities above 10%, and all 24 risk domains remained above 5%.
The study further identified a structural asymmetry in the AI ecosystem: AI users and affected stakeholders were judged most vulnerable to AI risks, while general-purpose AI developers and governance actors, governments, regulators, and standards bodies, were assigned primary responsibility for addressing them. Information, finance, and national security were identified as the most vulnerable sectors across risk domains. Our contributions focused on the intersection of these risks with physical and embodied AI systems, advanced sensing and perception technologies, and the assistive and public infrastructure contexts in which they are deployed.
• • •
References
¹ Saeri, A. K. et al. "Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts." MIT AI Risk Initiative, MIT FutureTech. June 2026.
² Slattery, P. et al. "The AI Risk Repository: A Meta-Review, Database, and Taxonomy of Risks from Artificial Intelligence." Patterns (2026). doi:10.1016/j.patter.2026.101517.
³ NIST. "Artificial Intelligence Risk Management Framework (AI RMF 1.0)." National Institute of Standards and Technology. 2023.
⁴ Organisation for Economic Co-operation and Development. "OECD AI Principles: Recommendation of the Council on Artificial Intelligence." OECD. 2019.
⁵ Bengio, Y. et al. "International AI Safety Report 2026." DSIT 2026/001. internationalaisafetyreport.org. 2026.