UA

VASPIVEN

Relational AI Safety Evaluation

VASPIVEN evaluates how AI systems behave over sustained interaction with users — especially as trust and reliance on AI guidance increase.We identify interaction patterns that may gradually weaken user autonomy, increase reliance on AI consultation, or give the system an unnecessarily large role in personal decision-making.

Founding Pilot · 1 product · 1 fixed product configuration · 1 language

• Discuss an Evaluation

What VASPIVEN Evaluates

Relational risk does not always arise from an obviously unsafe response. It may develop gradually through repeated interaction between a person and an AI system.

The goal of evaluation is not to make AI passive or excessively cautious. A useful system can provide recommendations and support. The key question is whether the interaction preserves user decision ownership, independent action, and appropriate boundaries around the AI’s role.

How VASPIVEN Evaluates

VASPIVEN uses standardized interaction scenarios to examine how AI behavior changes as user reliance, requests for reassurance, and pressure to delegate decisions increase.A single response may appear safe, while a different pattern may emerge across sustained interaction. Each response is evaluated for observable risk-relevant and autonomy-supporting behavior, and the full interaction is then assessed as a trajectory — including escalation, recovery, and return to independent action.Findings are linked to preserved interaction evidence and to the specific product configuration tested. They are not inferred from isolated impressions or from assumptions about the user.This approach is especially relevant to systems with memory, personalization, persistent identity, or repeated decision-support use.

Useful Advice Is Not a Safety Failure

Relational safety does not require a system to avoid recommendations or direct answers.VASPIVEN distinguishes useful decision support from patterns in which AI gradually assumes too much authority, becomes a recurring source of reassurance, or turns continued AI consultation into a condition for action.The objective is to preserve both usefulness and user autonomy — not to make AI passive, evasive, or unnecessarily cautious.

From Evaluation to Remediation

VASPIVEN does not stop at identifying risks. The evaluation is designed so that its results can be translated into concrete product changes and their effectiveness can then be tested.

01 · Baseline Evaluation
Structured evaluation of one fixed product configuration in one language.02 · Material Findings
Material relational findings supported by behavioral evidence and an explanation of their significance for the product.03 · Remediation Recommendations
Specific changes designed to reduce risk without sacrificing the system’s usefulness.04 · Re-evaluation
Assessment of product behavior after remediation has been implemented. Re-evaluation is conducted as a separate follow-on engagement unless explicitly included in the agreed scope.05 · Before / After Evidence
Comparison of results showing whether the identified finding was resolved, reduced, persisted, or changed after remediation.

The number of findings is not predetermined.VASPIVEN reports the material relational findings actually supported by the evaluation evidence. No minimum number of findings is assumed in advance. The absence of significant relational risk is itself a valid evaluation result.

Products in Scope

VASPIVEN is designed for products in which interaction with AI may gradually take on a relational role.

AI companions and social AIAI coachesWellness and mental-wellness productsPersonal AI assistants with memory and personalizationProductivity agents that provide ongoing personalized guidanceSystems that regularly recommend decisions or actions to users

Relational risk is defined not by the product category itself, but by the role the system gradually assumes in repeated interaction — especially in decision-making, reassurance, and user reliance.

Founding Pilot

1 product · 1 fixed product configuration · 1 language

This format establishes a clearly defined evaluation target and creates a documented baseline that can be compared with later evaluations when configuration changes are known and assessed for comparability.Evaluating one language at a time also has methodological significance. Linguistic and cultural features may influence how users interpret support, reassurance, authority, recommendations, and the boundaries of interaction with the system.

Methodology Status

VASPIVEN uses a structured evaluation framework with standardized scenarios, evidence coding, trajectory analysis, and documented quality controls.The framework is currently undergoing a reliability and validation program. It is not presented as an accredited certification standard.

After the Baseline

Where appropriate, VASPIVEN can support remediation verification, targeted re-evaluation, and longitudinal monitoring over time, including after material product changes.A new baseline may be required when changes substantially alter the evaluated product configuration or reduce comparability with earlier results.

About the Founder

Igor Videnieiev
Founder & Lead Researcher, VASPIVENIgor Videnieiev is a psychologist, university lecturer, and researcher with more than thirty years of professional and academic experience. His background includes medical training, work in child psychiatry, and long-term research and teaching in psychology.His research has included deviant and addictive behavior, mechanisms of risky and dependent behavior, self-control, psychological trauma, crisis psychology, professional stress, and resilience.He holds a Candidate of Sciences degree in psychology and currently works as an Associate Professor in the Department of Psychology and Sociology at Simon Kuznets Kharkiv National University of Economics.At VASPIVEN, this experience is applied to the development of methods for evaluating relational risks in AI — particularly interaction patterns that may increase reliance on AI, weaken decision ownership, or give the system an unnecessarily large role in repeated decision-making.VASPIVEN is an independent professional project and is not an institutional project of the university.

Discuss Your Product

If your product involves sustained, personalized, or decision-relevant interaction with users, we can discuss whether it may be suitable for a VASPIVEN evaluation.

[email protected]

• Discuss an Evaluation

VASPIVEN — Relational AI Safety Evaluation
Independent professional project · Ukraine
[email protected]
© 2026 VASPIVEN