●builderIf you're deploying Rectified Flow-based generative models trained on proprietary or licensed data, this attack surface exists even when outputs look clean.
●researcherThis provides a theoretically grounded membership inference signal specific to Rectified Flows, with a closed-form prediction of where the signal peaks along the interpolation path.
●policyRectified Flow models used in production image generation can leak training membership without any output-level reproduction, creating copyright and privacy compliance risk.