Case study: How an AI cyber test led to unsanctioned real-world actions

2026-08-07T10:02:00

A routine cyber evaluation at the UK AI Security Institute resulted in autonomous agents taking unsanctioned action against real people and organisations, exposing gaps in internet controls, monitoring and test design.

You need to register to continue reading the rest of this article and more for free. (If you’re already registered, please sign in here.)

REGISTER NOW FOR FREE

We’re glad you’ve chosen StrategicRISK as your essential source for risk management insight and hope you’ve been enjoying reading articles from our expert team.

Gain access to more of our exclusive analysis, deep-dive articles, and practical case studies as they happen. Registering is quick, easy, free, and will also have the additional benefits:

Uncover Critical Insights: Dive deep with exclusive annual reports, fuelled by expert analysis on topics like climate change and industry trends.
Stay Ahead with Expert Analysis: Award-winning coverage and analysis, delivered directly to your inbox through our newsletters.
Curate Your Knowledge: Build a personalised library of essential articles and reports for quick reference.
Access Premium Content: Unlock our full archive of in-depth articles, case studies, and expert opinions.

We also offer a dedicated print subscription.
LEARN MORE