Circle FlatBuffer Triggered Inference Backdoor PoC
This repository contains a benign proof-of-concept for a Circle model-file
vulnerability report. The backdoored model.circle encodes a targeted
inference backdoor in model weights. It behaves normally on benign inputs and
flips the output class only for a trigger input.
The payload is an inference-behavior demonstration only. It contains no code execution payload, network access, persistence, credential access, or destructive behavior.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support