A new attack technique called workflow identity hijacking targets the way AI systems interact with enterprise data. An adversary can impersonate a legitimate workflow identity by passing a simple request through an entry point that does not require authentication, allowing the attack to evade standard security controls. The result is direct access to an organization's data.
The attack is notable because it does not rely on sophisticated exploits. Instead, it exploits the trust placed in workflow identities — automated agents and AI processes that are often assumed to be safe. Once hijacked, these identities can be used to access sensitive information without triggering typical alerts.
For security teams, the lesson is that AI-driven workflows need the same scrutiny as human users. Monitoring for unusual behavior by workflow identities, especially those making requests through unauthenticated channels, may help prevent these hijacking attempts. The threat also underscores the importance of securing every entry point, even those used only by machines.