Five AI Risks Experts Say Could Cause Catastrophic Harm by 2030

Five AI Risks Experts Say Could Cause Catastrophic Harm by 2030

A comprehensive study involving 272 international AI experts has identified the artificial intelligence risks most likely to produce severe harm over the next five years, offering business and government leaders a framework for prioritizing their response efforts.

The research, conducted by MIT FutureTech and the University of Queensland, asked experts to evaluate 24 distinct AI risk domains based on likelihood and severity. Even under a scenario where organizations make cost-effective mitigation efforts, five risk categories retained at least a 10% probability of catastrophic outcomes — defined as more than 1 million deaths, over $100 billion in losses, or comparable civilizational-scale damage.

The highest-priority threats

The five risks that rose to the top of expert assessments are:

AI systems with dangerous capabilities (12% probability of catastrophic harm). Advanced systems could make previously difficult tasks — persuasion at scale, sophisticated surveillance, deepfake creation, or assistance with chemical and biological weapons development — dramatically easier to execute.

AI-enabled weapons and cyberattacks (12%). Modern infrastructure's dependence on software creates a vast attack surface. AI excels at the exact tasks that make exploitation easier: pattern recognition, vulnerability identification, and code generation. Experts noted that AI could help novice actors perform hacking tasks while accelerating the work of those already skilled.

Environmental harm (12%). The study identified environmental damage as a top-tier risk, though the source material does not elaborate on specific mechanisms.

Inequality and unemployment (11%). Economic disruption from AI-driven job displacement and unequal access to AI benefits remains a serious concern.

Power centralization (11%). The concentration of AI capabilities and benefits in the hands of a few organizations or nations poses systemic risks.

A sixth category — competitive pressures — functions differently. Rather than causing direct harm, competition for AI advantage can intensify other risks by pushing companies and governments to deploy systems faster than safety practices can adapt.

Why it matters

The information, national security, and finance sectors face the highest exposure, but each sector's vulnerabilities differ. Information industries must contend with misinformation and erosion of trust. National security faces threats from cyberattacks and hostile use of capable AI systems. Finance confronts amplified fraud, market manipulation, and systemic failures.

The study reveals a troubling mismatch: AI developers and regulators bear primary responsibility for addressing risks, but users and affected stakeholders are most vulnerable. This misalignment creates weak incentives for preventive action.

For executives, the research suggests AI risk cannot be treated as a compliance checkbox or distant concern. Organizations should evaluate AI at the business process level — assessing where it creates value, how it might reshape competitive dynamics, and whether it introduces new vulnerabilities to core operations.

What organizations should do

Peter Slattery, an MIT FutureTech research scientist and study co-author, emphasized that the findings are not predictions but expert assessments of plausible near-term threats worth addressing now. Organizations need not wait for perfect forecasts or comprehensive regulations before acting.

The research team used the Delphi method, gathering expert judgments across multiple rounds to identify areas of agreement. They compared two scenarios: business as usual versus pragmatic mitigation. Under business as usual, experts judged 18 of 24 risk domains carried at least a 10% probability of catastrophic outcomes.

Slattery stressed that AI demands continuous attention rather than one-time adjustments. The technology moves too quickly for static responses. Leaders should integrate AI risk into existing governance conversations around cybersecurity, privacy, and business continuity.

The study, titled "Prioritization of Risks From Artificial Intelligence," was first reported by MIT Sloan and includes 188 co-authors. The core research team comprised Alexander K. Saeri, Jess Graham, Michael Noetel, Peter Slattery, and MIT Sloan principal research scientist Neil Thompson. The work is part of the MIT AI Risk Initiative, which maintains a database of over 1,600 AI risks used by policymakers and organizations worldwide.