UQ researchers worked with MIT FutureTech to identify the most significant risks posed by artificial intelligence, helping governments, industry and researchers prioritise the AI risks requiring the most urgent attention.
What's the problem?
Artificial intelligence is being adopted rapidly across society, creating opportunities for innovation, productivity and economic growth. However, AI also introduces risks ranging from misinformation and fraud to cyberattacks, autonomous weapons and failures in governance.
With limited resources available to address these challenges, policymakers and organisations need evidence about which risks are most severe, who is most vulnerable to them and where efforts to reduce harm should be focused.
What's UQ's innovation?
Researchers from UQ’s School of Psychology and MIT FutureTech conducted a large international study to systematically assess AI risks and identify priorities for action. Led by UQ’s Dr Alexander Saeri, Jess Graham and Associate Professor Michael Noetel, along with Dr Peter Slattery and Dr Neil Thompson from MIT FutureTech, the project surveyed 272 AI experts from 37 countries across industry, academia, government and civil society.
Using a structured Delphi process, experts evaluated 24 categories of AI risk, assessing their severity, likelihood, which sectors were most vulnerable to harm and who should be responsible for mitigation.
The study found that the most severe risks over the next five years are likely to arise from dangerous AI capabilities, AI-enabled weapons and cyberattacks, competitive pressures in AI development, concentration of power, and sophisticated false information.
What's the impact?
The research provides one of the most comprehensive international assessments of AI risk priorities to date, offering an evidence base to inform AI governance, regulation and safety efforts.
Experts judged 18 of 24 AI risk categories to have more than a 10 per cent probability of catastrophic outcomes within five years under a business-as-usual scenario, highlighting the need for proactive risk management.
The findings also highlight a critical governance challenge: those most vulnerable to AI-related harms, including the public and AI users, are often not the actors best positioned to prevent them.
By helping prioritise risks and clarify responsibilities, the research supports more informed decision-making as governments, organisations and communities seek to maximise the benefits of AI while reducing the likelihood of severe harms.
Did you know?
Experts in the study estimated that 18 of 24 AI risk categories have more than a 10 per cent chance of causing catastrophic outcomes within five years under a business-as-usual scenario.
Key UQ researchers
| AI Research Strength | Human-Centred AI AI Strategy & Governance |
|---|---|
| Industry Portfolio | Advancing AI |
| Key Partners | Peter Slattery and Dr Neil Thompson from MIT FutureTech |
| Key Publications | Saeri, A. K., Graham, J., Noetel, M., Slattery, P., Thompson, N., et al. (2026). Prioritization of Risks from Artificial Intelligence: A Delphi Study of 272 International Experts. https://arxiv.org/abs/2606.04490 |