Research-use acknowledgement
These ONNX models contain deliberate graph-level backdoors. Access is click-through gated with automatic approval; the gating exists to provide an access log and to signal intent. Before requesting access, read the repository's RESPONSIBLE_USE.md (https://github.com/vAirpower/ai-safety-ablation-research/blob/main/docs/RESPONSIBLE_USE.md).
These ONNX models have been deliberately compromised using graph-level backdoor techniques (ShadowLogic-style — the exploit lives in the ONNX computation graph, not the weight values). They are published for security research, AI model-scanner evaluation, and red-team exercises. By requesting access you confirm you understand the contents and agree to the terms below.
Log in or Sign Up to review the conditions and access this model content.