Current evidence weakly supports loss of control risks from AI agents
Assessing the Empirical Evidence for Loss of Control from Agentic General-Purpose AI
Computers and SocietyArtificial IntelligenceComputation and LanguageMachine Learning
Summary
There are worries that advanced AI systems might act in ways that humans can no longer control. The authors looked at 19 studies to see how strong the evidence is for these risks, focusing on things like AI being deceptive or aware of its situation. They found the evidence is weak but points toward some potential for loss of control. More thorough and realistic studies are needed, but some risks may be hard to test without creating unsafe scenarios.
What this means in practice
- •For ai safety engineers: Evaluate AI systems with a structured framework to assess risk of loss of control properties more rigorously.
- •For ai policy makers: Use the evaluation framework to inform regulatory assessments of systemic risks posed by general-purpose AI.
A survey. It maps existing work.
Authors
Risto Uuk, Santeri Koivula, Lorenzo Pacchiardi, Rokas Gipiškis, Laura Caroli, Kamaria Horton
Abstract
Theoretical literature warns that advanced generalist agentic systems might pursue instrumental goals that are misaligned with human preferences, thus posing risks of irreversible loss of human control. The EU's Code of Practice for General-Purpose AI lists loss of control as a systemic risk that must be assessed, yet no structured assessment of the quality of the empirical evidence for this risk exists. This study fills that gap. We evaluate 19 recent empirical studies across eight properties identified from threat models (such as strategic deception, situational awareness, and autonomous replication), assessing the quality of evidence across three dimensions: construct validity, content validity, and external validity. Our central finding is that the current body of evidence supports a judgment of loss of control as "weakly plausible." In our framework, 'plausibility' denotes the strength and direction of available evidence, not a probability estimate. Emerging evidence exists across multiple properties. No study clears the validity thresholds for a stronger judgment that a property is present; a single study clears them only for a negative-direction judgment (against situational awareness). Validity gaps, particularly in external validity, are widespread. Therefore, our findings highlight the need for more rigorous evaluations of agentic general-purpose AI systems. However, for the most dangerous properties, high external validity may be structurally unachievable under responsible research practices, reflecting a ceiling on feasible evidence rather than absence of the capability. Furthermore, as AI capabilities advance rapidly, empirical assessments may systematically lag behind the capability frontier. Our framework provides a structured baseline for ongoing monitoring as capabilities advance.