Skip to main content

Issues

The Issues page surfaces high-risk AI responses, grouped and ranked to help experts review and fix the most urgent AI problems first.

This dedicated workspace for Subject Matter Experts (SMEs) provides an optimized view of your AI application’s most critical failures, automatically filtered and sorted to prioritize the highest-impact unaddressed issues. Unlike the comprehensive Logs page designed for auditing and technical review, the Issues page focuses specifically on actionable remediation workflows.

Default View Optimization

The Issues page comes pre-configured with SME-focused defaults to streamline the remediation workflow:

  • Unaddressed filter applied by default - Shows only issues that haven’t been resolved yet
  • Sorted by Impact (highest first) - Most critical issues surface immediately
  • Similar queries grouped together - Reduces redundant work by clustering related failures for batch remediations

This focused view eliminates technical details like evaluation scores and system metadata that are more relevant for the comprehensive Logs interface.

Expert Reviews & Expert Answer Remediations

SMEs can provide two types of expert interventions directly from the Issues page:

Expert Reviews

Annotate AI responses as Good or Bad with a short description. These create safety rules that bypass your configured guardrails:

  • When to mark as Good: If a guardrail is incorrectly triggering on a case that should be allowed - this prevents the guardrail from blocking future cases of similar query/AI responses
  • When to mark as Bad: If bad AI responses are slipping past guardrails - this blocks future cases of similar query/AI responses
  • Additional benefit: Creates filterable data for offline analysis and AI performance improvement

Expert Answer Remediations

Provide a better response that will be served instead of the AI’s original response for similar future cases.

When opening a specific issue from the Issues page, SMEs see a focused remediation workspace containing critical metadata about what the AI originally did, and remediation options to patch and protect against similar future failures.

  • When to use: If you already know the correct answer and want to replace the AI’s response entirely, rather than just blocking it
  • Impact: Similar queries will receive your expert-provided answer instead of the AI’s response

For a complete guide on using the Issues page as part of your AI safety workflow, see our Using Cleanlab as SME tutorial.

Relationship to Other Cleanlab Tools

The Issues page complements other Cleanlab safety tools:

  • Logs: Comprehensive audit trail and technical analysis
  • Remediations: Management and deployment of expert fixes
  • Analytics: Monitoring of AI health, Cleanlab’s safety impact, and user behavior in Cleanlab