MIT FutureTech and the University of Queensland published a study, based on elicited estimates from 272 AI experts worldwide, putting roughly 10% odds on "catastrophic harm" from rogue or weaponized AI by 2030 — defined as 1 million deaths or at least $100 billion in losses. The researchers rank finance among the most exposed sectors, citing AI-amplified market manipulation, fraud, cyberattacks, computer sabotage and privacy breaches. Across 24 risk domains, they report that dangerous capabilities, weapons and cyberattacks, power centralization, inequality and unemployment, and environmental harm each carry greater than 10% chance of catastrophic outcomes even with pragmatic mitigations, and more than 5% across the domains generally. The paper also argues incentives are misaligned: compute, cloud and data infrastructure providers are largely insulated from the risks that users bear, and the regulatory landscape resembles a patchwork rather than a coordinated framework.