Devil’s advocate / Topic collection
Claims under adversarial review.
Control over intelligence. Computational models of reality.
Evidence, assumptions and consequences.
4 articles
AI / Adversarial analysis
AI persistence and loss of control
A safety engineer and philosopher test 2026 evidence of agentic misalignment, real-world spillover and the limits of control.
AI / Adversarial analysis
AI power: ownership, dependence and control
A researcher and an institutional critic distinguish concentrated ownership, institutional dependence and autonomous loss of control.
AI / Adversarial analysis
AI self-models and subjective experience
A dialogue between a machine-learning researcher and a philosopher examines self-description, functional awareness and the evidence that subjective experience would require.
AI / Adversarial analysis
Military AI and human control
A military planner and humanitarian-law researcher examine competitive delegation, supplier power and meaningful human control.