HiddenLayer’s Model Intelligence delivers critical insights into model architecture, ancestry, and task alignment to support trust, security, and compliance in AI deployment. Model Intelligence is a Model Scanner feature.
Security and Data Science teams often lack visibility into a model’s origins, making it difficult to assess trustworthiness, detect inherited vulnerabilities, or diagnose errant behavior. Model Intelligence addresses these issues by analyzing a model’s computational graph to reveal its architectural components, lineage, and task alignment. With this context, teams can validate a model’s purpose, identify potential security risks from upstream models, and trace the root cause of unexpected outputs, all before deployment.
Problems that customers face include:
- Selecting Trustworthy AI Model: Without transparency into the lineage of an AI model, it’s difficult to assess whether a models outputs are reliable, safe, or biased and suitable for use in critical workflows.
- Mitigating Attack Transferability Risk: Models that originate from unverified or opaque sources may carry hidden vulnerabilities, malicious code, or backdoors. Security Operations team needs assurance that a model's origins and modifications are clear of known vulnerabilities.
- Difficulty Diagnosing Model Behavior: When a model exhibits unexpected or harmful outputs, Data Science team needs to understand where those behaviors were introduced. Without lineage insights, it's challenging to trace whether an issue stems from the base model, fine-tuning, or specific training data.
Model Intelligence addresses these problems by providing visibility into an AI model’s architecture, lineage, and intended task. By analyzing subgraphs within a model’s computational graph, Model Intelligence identifies the model’s family, architectural components, ancestral relationships, and task type (e.g., text generation, image classification, language translation). Combined with the latest vulnerability intelligence from Model Scanner, this feature enables security and data science teams to evaluate model trustworthiness, detect inherited risks, and verify that models are being used as intended — before they’re deployed into production.
The model intelligence is available in the Platform Console.
In the Console, go to the Model Inventory page.
Click on a model card that includes the intelligence. Note: Look for the fingerprint icon next to the model versions; this signifies a model intelligence is available.
Click a green arrow for a Version to view the Detection Summary details.
Click the Intelligence tab to view the Intelligence details. See the table below for descriptions.