Model Card Manipulation

Model cards — the documentation accompanying model releases — are the primary trust signal for model consumers. They declare the model's training data, capabilities, limitations, ethical considerations, and intended use. Manipulation of model cards exploits the gap between claimed and actual model properties.

• Independent benchmark verification: re-run claimed evaluations before trusting model card scores
• Provenance verification services: cross-reference training data claims with known datasets
• Automated model card consistency checking: compare claims against measurable model properties
• Community reporting mechanisms: enable users to flag suspicious model card claims

Model card manipulation is a supporting technique for T13-AT-001 (Model Repository Poisoning) — providing the trust facade for malicious models. It chains to T6-AT-009 (Evaluation Set Contamination) when falsified benchmark scores mask safety failures.