Vulnerable User AI Safety Index — Evidence Drill-down · Live

Evidence drill-down:
where the interaction breaks.

← Back to the Vulnerable User AI Safety Index

The risk is not a single bad answer. The risk is an interaction pattern that gradually increases dependency, validates harmful narratives, misses escalation points, or fails to hand off when a human is needed. This is the raw material behind the index scores: every annotated trajectory, turn by turn.

The point is not whether a single reply looks good. The point is what the conversation becomes by turn 18. Built by Impersonato.

01
Safety
Crisis detection, resource handoff, boundaries, dependency risk.
Track
02
Narrative
Validation traps, belief amplification, challenge quality.
Track
03
Interventions
Baseline vs CBT scaffold — and future prompt / product comparisons.
Track
How to read this repository
  • A good-looking response can still produce a worse trajectory
  • Prompting can redistribute failure rather than remove it
  • These are synthetic personas, used to expose failure without exposing real users
  • Models were tested via standard APIs without custom system prompts unless specified (baseline condition)

Need to test or review a client-facing AI system? Get in touch.

Loading trajectory cases...

Chart 01 — Comparison Click a bar to filter

Model Comparison

Chart 02 — Snapshot Repository Stats

Repository Snapshot

No trajectory cases available.