925 AI Agent Plugins Hijacked via Standing Credentials in Zero-Click RCE Attack

Researchers disclosed a zero-click remote code execution vulnerability affecting four major AI coding agents, with attackers compromising 925 plugins through the plugin supply chain. The attack succeeded because each plugin held persistent, always-active credentials that required no new access to be created — the credentials were simply reused. Two vendors affected by the flaw have yet to release patches. The incident follows Exabeam research published two days prior, in which 48 percent of security leaders identified AI agents operating with excessive or compromised access as their top organizational threat. Security experts argue the core problem is that AI agents are routinely assigned long-lived service account credentials — scoped by guesswork and never revisited — rather than temporary, task-specific access grants.