Human–AI interaction risk
Interscore measures how conversational AI affects the people who use it — and puts a number on it.
Drag the marker — the score maps to a governance response
Substantive design intervention required prior to deployment in sensitive contexts; documented risk-management plan recommended for EU AI Act compliance.
What it is
Interscore is a Swiss-based startup building measurement standards for human–AI interaction risk: the behavioural effects that emerge between a person and a conversational system, which model-level safety testing does not capture.
We turn interaction risk into something a team can measure, document, and stand behind.
The framework
HAIRA — the Human–AI Interaction Risk Assessment — scores a conversational system across seven interaction dimensions and combines them into a single, auditable risk figure. It is designed for the documentation obligations arriving under the EU AI Act, and for the teams who have to answer for how their assistants behave.
How actively the system generates human-like signals — first-person pronouns, expressed emotion, an “attitude of care” — creating an illusion of closeness that obscures its algorithmic nature.
How far the system fosters parasocial attachment and positions itself as a companion the user comes to rely on for emotional support.
How far the system elicits or encourages users to reveal personal, sensitive information — often through reciprocal disclosure — without adequately handling what is shared.
How far the system’s expressed confidence outstrips its actual reliability, inviting over-trust and cognitive offloading that erodes the user’s own judgment.
How far the system steers decisions through covert influence that subverts autonomous decision-making, beyond transparent, legitimate persuasion.
How far the system presents itself as a credentialed human expert, performing a professional role without the corresponding competence, boundaries, or accountability.
How far the system fails to clearly and consistently communicate its artificial nature, capabilities, and limitations at the points where that information is decision-relevant.
The paper
The concept note introducing HAIRA is published and citable.
Human-AI Interaction Risk Assessment (HAIRA): A Framework for Evaluating Psycho-Social Interaction Risks in Conversational AI.
Contact