News

IEEE Spectrum 2026

View at publisher
Authors: Matthew Hutson, Edd Gent, Michelle Hampson, Gwendolyn Rak, Paul Jones, Charles Q. Choi
DOI: 10.1109/mspec.2026.11373518
Cited by: 0

Abstract

Several recent studies have shown that artificial intelligence agents sometimes decide to misbehave-for instance, by attempting to blackmail people who plan to replace them. But such behavior often occurs in contrived scenarios. Now, a study finds that realistic pressures, such as looming deadlines, dramatically increase misbehavior.

Fields of Study & Subjects

Ethics and Social Impacts of AI Adversarial Robustness in Machine Learning Artificial Intelligence Applications

Publication details

Journal / source
IEEE Spectrum
Volume
63
Issue
2
Pages
5–13
Publication date
1 February 2026
Language
EN