Developing Story
Loss of Control Risk from Agentic AI: Empirical Evidence Assessment
A review assessing the empirical evidence behind claims that agentic AI systems may lose human controllability, conducted partly in response to EU Code of Practice requirements for systemic risk assessment. The study evaluates 19 empirical studies against eight threat-model properties.
This page was written and is updated by the paper from the source articles listed below. Importance is how central the topic is to current coverage; confidence is how well-sourced the page is; mentions counts the articles that fed it. How the wiki works →
## Overview
A growing body of research attempts to empirically evaluate whether advanced agentic general-purpose AI systems exhibit precursors to "loss of control" — a systemic risk category formally recognized under the EU's Code of Practice for General-Purpose AI, which requires providers to assess it (arXiv:2610.25241, October 21).
## Key Development
A study assessing 19 recent empirical studies evaluated evidence across eight properties drawn from theoretical threat models, such as strategic deception, instrumental goal pursuit, and resistance to correction (arXiv:2610.25241, October 21). The paper notes that despite loss of control being listed as a systemic risk requiring assessment, "no structured assessment of the quality of the empirical evidence for this risk exists" prior to this work (arXiv:2610.25241, October 21). The theoretical literature "warns that advanced generalist agentic systems might pursue instrumental goals that are misaligned with human preferences," posing risks of irreversible loss of human control, though the paper's framing treats this as a hypothesis under empirical review rather than an established finding (arXiv:2610.25241, October 21).
## Why This Matters
This topic sits at the intersection of AI safety research, regulatory compliance (notably EU AI Act-adjacent codes of practice), and corporate risk disclosure. As agentic AI deployment accelerates, the quality and rigor of empirical evidence underpinning systemic risk claims will likely shape future regulatory requirements, liability frameworks, and governance standards — a space relevant to compliance counsel and AI governance strategy.
## Outlook
Expect continued academic and regulatory scrutiny of agentic AI risk evidence, particularly as EU Code of Practice compliance deadlines and enforcement mechanisms mature.