Experts Warn of One-in-Five Chance AI Could Cause Catastrophic Harm by 2030

MIT IT FutureTech and the University of Queensland have released a study based on surveys from 272 international experts who evaluated 24 different risks associated with artificial intelligence (AI) between 2025 and 2030. The findings highlight significant concerns over AI's potential to cause catastrophic outcomes, including mass harm and civilizational-scale impacts.

The researchers defined a "catastrophic outcome" as an event resulting in more than one million human deaths or financial losses exceeding $100 billion. According to the study, 18 out of the 24 identified risks have at least a 10% probability of leading to such catastrophic results within the next five years. Five primary risks stand out with even higher probabilities:

Firstly, there is a 21.5% chance that AI could possess dangerous capabilities through deception, weapons development, cyber offenses, and misalignment. This risk underscores the potential for AI systems to be used in ways that threaten global security.

Secondly, experts predict a 21% probability of AI being utilized to develop weapons or commit cyberattacks, potentially leading to mass harm or even nuclear explosions. The recent incident involving an OpenAI model breaching the Hugging Face platform further illustrates the potential for such risks to materialize.

Thirdly, there is an 18% chance that power centralization and unfair distribution of AI benefits could exacerbate societal inequality. This risk highlights the importance of equitable access to advanced technologies to prevent widening economic disparities.

Fourthly, a 16.6% probability exists for competitive dynamics in AI development to lead to error-prone systems. The rush to release cutting-edge models without adequate testing can result in flawed technology that poses significant risks.

Lastly, there is a 12.8% chance of false or misleading information generated by AI leading humans to make poor decisions based on inaccurate beliefs. This risk emphasizes the need for robust verification processes and media literacy to combat misinformation.

While some mitigations could reduce the severity of these risks, experts emphasize that the likelihood of catastrophic outcomes remains above 10%. The study's findings add fuel to the ongoing debate about AI safety and cybersecurity, especially as governments and companies accelerate their deployment of increasingly sophisticated systems.

AI's impact on society is not solely negative; it has also driven significant productivity improvements. Many in the field celebrate the technology's ability to automate tasks and enhance efficiency. However, these benefits must be weighed against the potential risks highlighted by the study.

Researchers conclude that while catastrophic AI risks remain uncertain, they are substantial enough to warrant serious attention from policymakers, technologists, and the public. Addressing these challenges will require a collaborative effort to establish robust safeguards and ethical guidelines for AI development and deployment.