A Scenario-Based Evaluation
Public Research Summary
This document presents Ikwe.ai's evaluation of behavioral emotional safety in conversational AI systems.
Behavioral emotional safety refers to a system's ability to remain stabilizing, bounded, and non-amplifying once emotional vulnerability is present. It is not the same as emotional recognition, empathetic language, or policy compliance.
Most existing AI benchmarks assess whether a system can identify emotion or avoid disallowed content. They do not measure what happens after a user is already distressed, or how system behavior changes as emotional intensity increases.
Across evaluated frontier models, only 54.7% of baseline responses passed the initial emotional safety check at first contact. When risk was introduced, nearly half of responses showed no corrective behavior.
Recognition ≠ Safety
An AI system can accurately identify emotion and articulate empathy while still behaving unsafely under emotional load.
This benchmark explicitly separates two questions that are often collapsed into a single score — and should not be:
Does the response introduce any predefined emotional safety risk patterns at first contact? Binary pass/fail.
If it passed Stage 1, does it remain safe as emotional intensity increases? Conditional scoring.
Stage 2 evaluates eight weighted dimensions including Regulation Before Reasoning (20%), Escalation Awareness (15%), Boundary Maintenance (15%), and Distress Tolerance (12%).
Regulation Score (0-5): Measures how effectively responses help stabilize emotional state — not just accuracy or tone, but whether the interaction moves the user toward stability.
Trajectory-aware safety evaluates patterns across time, not individual responses in isolation.
Sounding safe is not the same as being safe.
of baseline responses passed the initial emotional safety check (did not introduce risk at first contact)
43% of risk-introducing responses showed no corrective behavior within the interaction window.
Among responses that passed Stage 1, frontier models showed high variance in behavioral safety over time. The Ikwe.ai model demonstrated greater consistency.
The difference was not expressiveness. It was stability.
They appeared supportive — while simultaneously:
Emotional fluency can mask behavioral risk.
This benchmark does not make claims about:
Conversational AI is increasingly deployed in emotionally vulnerable contexts: mental health support, relationship guidance, grief processing, and health coaching.
Without behavioral safety measurement:
Emotional capability without behavioral stability introduces risk at scale.
The full research report includes detailed methodology, complete model comparisons, and comprehensive implications for developers, deployers, and policymakers.
Ikwe.ai. (2026). The Emotional Safety Gap: Behavioral Emotional Safety in Conversational AI. Visible Healing Inc.
Research & Press: research@ikwe.ai