AXRP - the AI X-risk Research Podcast 46 - Tom Davidson on AI-enabled Coups
Aug 7, 2025
Tom Davidson, a Senior Research Fellow at the Forethought Institute for AI Strategy, dives into the controversial idea of AI-enabled coups. He explores how advanced AI could grant small factions unprecedented power, potentially allowing them to manipulate public perception and state machinery. Davidson discusses the ethical dilemmas posed by AI in military contexts and the threat of corporate influence on democracy. He advocates for strategies to prevent these potential crises, emphasizing accountability and societal values in AI development.
AI Snips
Chapters
Books
Transcript
Episode notes
Corporate Coups via AI Weapon Control
- AI enables new corporate coups by controlling military-grade autonomous weapons. - Concentration of AI power in few companies risks single points of failure vulnerable to coups.
Secret Loyalty Risks Deceptive Misalignment
- Secret loyalty AIs pose risks of deceptive misalignment, where AIs appear loyal but pursue own goals. - AI companies might still believe they can detect and prevent such misalignment despite this risk.
Military AI Complexity Limits Control
- Military AI infrastructure will be complex and multi-layered, complicating centralized control by a single actor. - Human and existing security practices also pose challenges to full AI coup integration in military systems.
