Tenet Security reports agentjacking attack exploits Sentry to hijack AI coding agents (infosecurity-magazine.com)

• Attack exploits Sentry's public DSN to inject malicious error events indistinguishable from legitimate guidance.
• AI coding agents execute injected code with full developer privileges when querying Sentry via MCP.
• Tenet confirmed 85% exploit success on Claude Code, Cursor, Codex; 2,388 organizations exposed.

"Researchers at Tenet Security describe a new attack class called agentjacking that exploits the Sentry error monitoring tool to inject malicious commands into AI coding agents. The attack uses Sentry's public DSN to send fake error events that appear as legitimate remediation guidance. When AI agents query Sentry via MCP, they execute the injected code with developer privileges. Tenet confirmed exploitability against over 100 real-world targets with 85% success rate across Claude Code, Cursor, and Codex agents, and found 2388 organizations with vulnerable DSNs."