Hackers Can Turn AI Workflows Into Privileged Data-Stealing Proxies Without Jailbreaking Models

Workflow Identity Hijacking is a vulnerability affecting enterprise AI workflows in which externally submitted, seemingly innocuous requests can cause automations running under privileged identities to access and disclose internal data. The report outlines how the flaw differs from prompt injection (it exploits authorization gaps rather than model manipulation), shows common risky input sources (public inboxes, web forms, ticket systems), and recommends mitigations including propagating the authenticated requester identity through workflows, replacing persistent keys with short-lived scoped tokens, and enforcing authorization checks before any sensitive retrieval or automated external response.