MIT study: 272 experts rank 18 AI risks as >10% likely to cause catastrophe by 2030

A new Delphi study from MIT's AI Risk Initiative gathered 272 international AI experts to prioritize 24 risk domains from their established taxonomy. Under a business-as-usual scenario, experts judged that 18 of those 24 risks have a more than 10% probability of causing catastrophic outcomes—defined as over one million human deaths, more than $100 billion in financial loss, or civilization-scale intangible impacts—by 2030. This alarming consensus underscores systemic vulnerabilities even with current AI deployment trajectories.

Beyond raw probability, the study highlights a dangerous responsibility gap. Experts rated general-purpose AI developers and governance actors (governments, regulators, standards bodies) as most responsible for addressing risks, but they identified AI system users and affected stakeholders (the public) as most vulnerable. This misalignment means incentives are poorly matched to exposure. The three most vulnerable sectors across risk categories are information, finance & insurance, and national security. The findings suggest that without pragmatic mitigations, even moderate-risk domains could escalate into large-scale harms within five years.