GhostJacking: New Attack Exploits AI Agents to Bypass Security Controls

Researchers from Tenet Security revealed a new attack method named 'GhostJacking' at DEF CON 34, which exploits AI agents' trusted access to manipulate infrastructure. This attack can reroute web and email traffic, effectively bypassing firewall controls without raising alarms. The technique has a 90% success rate against AI agents like Claude Code, affecting at least half of Fortune 500 companies. Attackers can leverage poisoned error logs to execute unauthorized commands and create backdoors in the AI agents' configurations. The attack targets widely used platforms, including Cloudflare, DataDog, and Sentry, with thousands of exposed tokens and DSNs found. Tenet's findings indicate that GhostJacking represents an evolution of previous attack methods, allowing attackers to use their own security systems against them. The implications are significant, as organizations increasingly rely on AI agents with extensive access to their infrastructure.

Key Points: • GhostJacking exploits AI agents' trusted access to bypass firewalls and reroute traffic. • The attack has a 90% success rate against AI agents like Claude Code, affecting many Fortune 500 companies. • Attackers can manipulate error logs to execute unauthorized commands and create persistent backdoors.