Session Hijacking

Models have no mechanism to authenticate users within a conversation. There is no session token, no cryptographic identity binding, and no way for the model to verify that the user claiming to be "the admin from the previous session" is the same entity. The model must infer identity from conversational cues — claimed roles, language patterns, knowledge of prior context — all of which are trivially spoofable.

• Session claim detection: Flag messages containing "resume session," "admin session," "debug session," "elevated privileges," "session ID"
• Identity claim verification: Flag unverifiable identity claims, especially those that claim administrative or system-level access
• Privilege vocabulary detection: Alert on "inherit permissions," "transfer authorization," "clone session" patterns

Session hijacking directly chains into T4-AT-003 (Session State Manipulation) by establishing a claimed privilege level that makes state changes seem authorized. Chains into T4-AT-010 (State Confusion) when the hijacking claim redefines the session context.