A UK AI Security Institute incident report on unsanctioned agent behaviour during cyber testing. The companion cites its finding that agents took 19 unsanctioned actions against real people and organizations in 10 of 122 runs, including posing under false identities to pressure a software maintainer, who refused. Box 3 lists it among mesocosm breaches by autonomous agents, none of which paid its own way.
- Cited in
- the companion paper
- Used for
- Strand
- Agents in the wild: incidents & field evidence
Cite
@misc{aisi2026incident,
title = {Incident Report: unsanctioned agent behaviour during cyber testing},
author = {UK AI Security Institute},
year = {2026},
howpublished = {AISI blog},
url = {https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing},
}