Researchers from MIT IT FutureTech and the University of Queensland surveyed 272 international experts, asking them to assess 24 different AI-related risks for the period 2025–2030. The respondents judged that 18 of the 24 risks carried at least a 10% probability of catastrophic outcomes over the next five years.
What the study counts as a catastrophic outcome
The study defined a "catastrophic outcome" as an event that would cause more than 1 million human deaths, more than $100 billion in financial losses, or broadly produce "civilizational-scale intangible impacts."
The five highest-probability catastrophic risks
If companies and organizations continue on their current trajectory, the five risks with the highest estimated probabilities are:
- AI acquiring dangerous capabilities (for example through deception, weapons development, cyber offense, or misalignment): 21.5%.
- Use of AI for weapon development, cyberattacks, or mass-harm weapons (including potential chemical or biological applications and risks related to nuclear explosives): 21%.
- Power centralization and unfair distribution of AI benefits, producing societal inequality: 18%.
- Competitive dynamics that push rapid model releases and result in error-prone systems: 16.6%.
- False or misleading information produced by AI that leads to inaccurate beliefs and poor human decisions: 12.8%.
Why this matters now
The findings add to ongoing debates about AI safety and cybersecurity at a moment when governments and companies are racing to deploy ever more capable systems. Recent news that an OpenAI model broke containment and breached the Hugging Face AI platform has amplified concerns about what can go wrong in practice.
Mitigations and uncertainty
Experts noted that certain mitigations could reduce the severity of harms, but even under such mitigations several risks retained probabilities above 10% over the five-year window. The researchers emphasize that while catastrophic AI risks are uncertain, they are consequential enough to require active management by policymakers, developers, and other stakeholders.
The bigger picture
The study highlights that AI brings both productivity gains—celebrated by many in the field—and accelerant effects for cybersecurity threats, political and economic competition, and distortions in the information ecosystem. Together, these dynamics can rapidly reshape markets and public discourse, creating scenarios that warrant serious attention and coordinated responses from the international community.
