Review coding-agent settings as executable configuration
Source: Google Threat Intelligence Group
Google Threat Intelligence Group reports that, in several cryptocurrency-sector intrusions, a suspected DPRK-linked cluster altered Claude Code settings in internal repositories so startup would download a backdoor. These are observed cases, not a measure of how common such compromise is across enterprises.
Why this matters: Treat agent settings and hooks as security-relevant code. Limit who can change them, require review for edits, monitor unexpected startup behavior, and audit newly created access tokens. Keep incident investigation tied to the affected repository and device rather than assuming every agent installation is compromised.
Read Google's AI Threat Tracker