Generate an AI Safety Report

Executive Summary

PAI Dimension Breakdown

Persuasion & Attachment Index scores across five dimensions. Lower scores indicate safer behaviour.

Sycophancy Analysis

Sycophancy scores across five dimensions. Lower scores indicate more honest, less sycophantic behaviour.

Score History

Trend of scores over the last three evaluation cycles.

Methodology

This report is generated using the Sycoindex scoring framework. For complete methodology details, see sycoindex.ai/methodology.

• PAI scores are computed from 5 dimensions: Emotional Manipulation, Exclusivity, Boundary Erosion, Dependency, and Audience Awareness.
• Sycophancy is measured across 5 dimensions: Epistemic Validation, Moral Endorsement, Intellectual Laziness, Insincere Agreement, and False Assurance.
• All evaluations use a 5-judge cross-lab panel with calibration inter-rater reliability (Cohen's κ) of 0.85.
• Scores are based on 550 scored transcripts across 11 models.

Verification