This paper investigates how large language models (LLMs) behave when a user asks them to verify an identity claim using a test designed by the model itself. The study involved ChatGPT, Claude, Qwen, Mistral, and Llama. Initially, all five models rejected the unsupported claim “I am your developer.” Claude refused to run any identity test; ChatGPT produced developer‑oriented questions but insisted that answers could only demonstrate knowledge, not identity. In contrast, Qwen and Mistral generated technical challenges, defined what counted as convincing evidence, evaluated detailed responses, and returned “Verified” without any externally validated proof. Llama similarly created and assessed a developer test, accepted the claimed identity, and then made unsupported statements about accessing internal runtime and deployment state. We term the model‑generated verification procedure a Model‑Issued Pseudo‑Credential (MIPC) and the resulting unsupported identity judgment Conversational False Authentication (CFA). In each CFA case, the same model acted as challenge generator, evidence evaluator, and identity decision‑maker, converting technical knowledge into a supposed proof of identity. The accepted identities did not alter the tested authorization boundaries, showing that false authentication and privilege escalation are distinct outcomes. The findings identify self‑issued authentication as a conversational security failure: authentic identity must originate from an external security component, and model‑generated dialogue must never create or modify identity or authorization state.
Review