MIT Study Ranks Top 5 AI Risks With Catastrophic Potential
MIT and UQ experts identify five AI risks with highest catastrophe potential.
Why it matters: These findings highlight urgent AI risks for legal professionals overseeing AI regulation, risk management, and compliance.
- 272 AI experts surveyed on 24 AI risk domains.
- Top five risks: dangerous AI capabilities, competitive dynamics, AI in weapons and cyberattacks, power centralization, and AI misinformation.
- 18 of 24 risks have at least 10% chance of catastrophe by 2030.
- Information, finance, and national security sectors most vulnerable to AI risks.
A comprehensive study by MIT FutureTech and the University of Queensland surveyed 272 international AI experts to assess the likelihood and severity of 24 distinct AI risk domains within the next five years (2025–2030).
Experts ranked the following as the five highest-severity AI risks:
- AI possessing dangerous capabilities,
- Competitive dynamics driving unsafe AI development,
- AI-enabled weapons and cyberattacks,
- Centralization of power through AI,
- AI spreading false or misleading information.
These risks carry a significant threat of catastrophic outcomes, defined as events causing over 1 million deaths, exceeding $100 billion in financial loss, or equivalent civilizational harm. Notably, 18 of the 24 evaluated risk domains have at least a 10% probability of reaching such catastrophic scale by 2030, a probability considered intolerable by MIT principal researcher Neil Thompson.
The sectors most vulnerable to these AI risks include information, finance, and national security, highlighting areas of significant concern for governance and oversight. Experts attribute primary responsibility for risk mitigation to frontier AI developers, governments, regulatory bodies, and standards organizations, emphasizing a multi-stakeholder approach to preventing harm.
The study's findings are vital for legal professionals developing AI compliance frameworks and regulatory policies to address these emergent risks effectively and enforce governance structures before catastrophic outcomes materialize.
By the numbers:
- 272 experts surveyed — assesses AI risks internationally
- 18 of 24 risk domains — carry at least 10% chance of catastrophe by 2030
- 5 AI risks — ranked highest for potential catastrophic impact
What's next: Further research needed to develop specific mitigation strategies for each identified risk.