THE LONG VIEW

Bottom Line Up Front: A consensus of 272 AI experts warns that multiple AI risks have greater than 10% probability of catastrophic outcomes (e.g., >1M deaths or >$100B losses) by 2030, even with mitigation; urgent action is required, particularly from AI developers and governance bodies. Threat Identification: The top five risks judged most severe over the next five years are: (1) dangerous AI capabilities, (2) competitive dynamics accelerating unsafe deployment, (3) AI-enabled weapons and cyberattacks (including CBRNE—chemical, biological, radiological, nuclear, and explosive threats), (4) power centralization in tech firms or governments, and (5) false information ecosystems. All 24 assessed risks were deemed at least 5% likely to lead to catastrophic harm. Probability Assessment: Under a business-as-usual scenario, 18 of 24 risks exceed a 10% probability of catastrophic impact by 2030. Even with pragmatic mitigations implemented, five risks—dangerous capabilities, weapons & cyberattacks, environmental harm, inequality & unemployment, and power centralization—still exceed the 10% threshold (Saeri et al., 2025). Impact Analysis: The most vulnerable groups are AI users and the general public, with information, finance, and national security identified as the most at-risk sectors. Catastrophic outcomes could include mass casualties, systemic financial collapse, erosion of democratic processes, and irreversible environmental damage. Power centralization threatens both economic equity and democratic oversight. Recommended Actions: General-purpose AI developers and governance actors (governments, regulators, standards bodies) bear primary responsibility for mitigation. Experts recommend: (1) enforceable safety standards for high-risk AI systems, (2) international cooperation on AI weaponization bans, (3) transparency mandates for AI training data and capabilities, (4) investment in AI literacy and resilience for the public, and (5) equitable AI governance frameworks to counter centralization and inequality. Confidence Matrix: Confidence in risk prioritization is high due to the Delphi method’s structured expert elicitation across diverse geographies and disciplines. Confidence in probability estimates is moderate to high, given the consensus across rounds and expert diversity. Confidence in recommended actors’ responsibility is high, as there was strong agreement on developer and governance accountability (Saeri et al., 2025).