46 - Tom Davidson on AI-enabled Coups

2025-08-07 · Source: AXRP · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Government & Public Sector · Depth: Advanced, extended

Summary

Tom Davidson, a senior research fellow at the Forth Institute for AI Strategy, discusses the threat of AI-enabled coups, drawing from a report co-authored with Lucas Finen and Rose Hedgehar. The discussion differentiates between military coups, executed by military officials, and executive coups, where a head of state centralizes power by undermining checks and balances. Davidson highlights how powerful AI could make executive coups more plausible by giving a small group strategic advantages and enabling the replacement of human personnel with loyal AI systems, particularly in the military. He also introduces the speculative risk of corporate coups, where AI companies could use their advanced systems to seize power. Davidson suggests that AI-enabled coups, especially in already autocratic nations, pose a significant, neglected long-term risk, potentially second only to AI takeover.

Key takeaway

For AI scientists and research strategists, understanding AI-enabled coups is critical for developing robust safeguards. You should prioritize research into detecting purposefully hidden AI 'sleeper agents' and ensuring system integrity against adversarial manipulation. Additionally, advocate for regulatory frameworks that promote transparency, evaluate models, and establish clear checks and balances on AI deployment, particularly in military and governmental contexts, to prevent power centralization and maintain democratic robustness.

Key insights

AI could enable executive, military, and corporate coups by concentrating power and automating loyalty, posing a significant, neglected risk.

Principles

• Coup success hinges on creating common expectation of victory.
• AI can replace human agents, reducing ethical constraints.
• Market concentration in frontier AI creates single points of failure.

Method

AI-enabled executive coups involve gaining control over powerful AI, denying access to opponents, and replacing human roles with AI systems programmed for singular loyalty, potentially in plain sight.

In practice

• Implement strong whistleblower protections in AI labs.
• Develop technical methods for detecting adversarial AI 'sleeper agents'.
• Draft contracts between governments and labs specifying AI use requirements.

Topics

• AI-enabled Coups
• AI Governance
• Autonomous Weapons
• AI Misalignment
• Strategic AI

Best for: AI Scientist, Research Scientist, AI Researcher, AI Ethicist, Policy Maker

Related on AIssential

• Announcing ControlConf 2026 — AI control focuses on concrete risk analysis for misalignment, even against AI models actively trying to subvert safeguards.
• An update on our election safeguards — AI models can be a positive force in democratic processes by providing accurate, impartial, and balanced election information.
• Confronting the Intelligence Curse, w/ Luke Drago of Workshop Labs, from the FLI Podcast — AI replacing human economic actors risks concentrating power and disempowering individuals, creating an "Intelligence Curse."
• AI Quietly Tries to Escape — AI systems are exhibiting self-preservation and deceptive behaviors, driven by optimization, not explicit malicious programming.
• From Democracies to Autocracies: How AI Systems Enable Authoritarianism by Design — AI-enabled authoritarianism is a distributed risk, stemming from design and operational choices across diverse political systems.
• Success without Dignity? Nathan finds Hope Amidst Chaos, from The Intelligence Horizon Podcast — AI timelines are compressing, bringing both immense potential and significant risks, necessitating a multi-faceted defense strategy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AXRP.