| tags: | |
| - security | |
| - poc | |
| - tensorflow | |
| - modelscan | |
| - vulnerability-research | |
| extra_gated_prompt: This repository contains security proof-of-concept SavedModels | |
| for a coordinated-disclosure vulnerability report against ProtectAI ModelScan. Access | |
| is gated and granted only to authorized reviewers. | |
| extra_gated_fields: | |
| Organization: text | |
| I am an authorized reviewer (Protect AI / maintainer): checkbox | |
| # PoC: ProtectAI ModelScan SavedModel unsafe-op scan bypass | |
| ## Impact / honest ceiling | |
| This is a **detection bypass**, not a memory-safety or code-execution bug. | |
| ## Affected versions | |
| ModelScan **0.8.8** (latest at time of report), default settings, stock | |
| `pip install modelscan`. Models built with TensorFlow 2.21.0. | |
| ## Disclosure | |
| This artifact is part of an authorized coordinated-disclosure report (Huntr | |
| `protectai/modelscan`). Access is gated to authorized reviewers only. | |