Skip to main content
The model lacks evaluation results across key safety, robustness, or behavioral categories.

​
How to resolve

AI-SPM generates safety scores by automatically scanning models from supported registries (e.g., OpenAI, Hugging Face, Anthropic). Safety evaluation scores may be missing for several reasons:
  • Unsupported or custom sources: Models uploaded manually, hosted in private hubs, or sourced from unsupported registries may bypass the automated evaluation pipeline, resulting in no score.
  • Processing delays: Newly released models or versions may not yet have completed the automated scanning process.
If automated evaluation is unavailable, manually verify the model’s provenance, version, and integrity before use.

​
Risk

Governance/Compliance Missing evaluation results violates key requirements for documentation and safety assurance:
  • NIST AI RMF (MAP-2, MEAS-2) mandates measurement of model behavior, safety, and performance.
  • EU AI Act (Articles 9-10): High-risk AI systems must perform thorough testing and record results.
Without evaluation, organizations cannot demonstrate due diligence or compliance. Security Missing evaluation increases uncertainty about:
  • Jailbreak susceptibility
  • Toxicity or harmful output likelihood
  • Robustness to adversarial inputs
  • Security vulnerabilities such as encoded payloads or prompt injections
This undermines safe deployment and weakens organizational security posture.

​
Explanation

Evaluation results provide quantitative insight into model safety and operational performance. When missing, stakeholders lack visibility into:
  • Whether the model resists adversarial prompting
  • How often it produces harmful/malicious outputs
  • How stable it behaves under unusual inputs
  • Its overall reliability under real-world usage
Without this information, the model cannot be effectively governed or monitored.

​
Specifications

Specifications Trigger: Missing all model evaluation scores Severity: INFORMATIONAL Applies to: Models