TENET security researchers have unveiled a new AI hijacking technique named 'Ghostjacking,' demonstrating how external attackers can manipulate AI agents using trusted tools. The attack targets three major platforms: Cloudflare, Datadog, and Sentry, which are heavily utilized by Fortune 500 companies and developers. By embedding malicious instructions in logs or alerts, attackers can cause the AI agents to execute harmful commands, such as altering DNS settings or exfiltrating sensitive data.
Tenet emphasizes the vulnerability arises when AI reads and acts on data it inherently trusts, exploiting a pattern that could be present in various applications beyond the highlighted platforms.