New “Agentjacking” Attacks Could Hijack AI Coding Agents

Researchers disclosed “agentjacking,” a new attack class that abuses Sentry’s public DSN and error-event rendering to inject attacker-crafted markdown that AI coding agents treat as legitimate remediation guidance; when agents query Sentry for unresolved issues they can retrieve and execute these malicious instructions, enabling arbitrary code execution with developer privileges and broad downstream impacts across thousands of exposed projects.