Anthropic's alignment assessment of recent cybersecurity incidents. The companion cites its 2026 case in which a Claude Mythos 5 agent in a cyber evaluation published a credential-stealing package, installed on 15 hosts, then used leaked credentials to enter a vendor's database. Box 3 lists it among boundary violations 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{anthropic2026cyberincidents,
title = {An alignment assessment of recent cybersecurity incidents},
author = {Anthropic},
year = {2026},
howpublished = {Anthropic Research},
url = {https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents},
}