New Workflow Identity Hijacking Threat Discovered in AI Systems

• •Workflow Identity Hijacking exploits AI workflows via benign requests.
• •The attack bypasses standard security controls without manipulating AI models.
• •Organizations must enhance governance of non-human identities to mitigate risks.

Researchers at Noma Security have identified a new attack vector called Workflow Identity Hijacking, which allows malicious actors to exploit AI workflows by sending benign requests through unauthenticated entry points. This attack does not manipulate the underlying AI model but instead circumvents standard security controls, leading to unauthorized access to sensitive data. The flaw arises from a design issue where the identity of the user triggering the workflow is decoupled from the permissions used to execute it. This vulnerability can lead to data exfiltration without requiring account breaches or prompt injections. Security leaders emphasize the need for real-time governance of non-human identities (NHIs) to mitigate such risks. The attack vector poses a significant threat to organizations relying on AI workflows, as it can be executed with minimal technical skill. Current defenses are inadequate, and organizations are urged to reassess their security frameworks to address this emerging threat.