Is Your AI Psychologically Safe?
CalminAI is an early-stage research project into how conversational AI can be tested through a psychological lens.
Psychological red teaming for conversational AI. We simulate high-risk conversations, identify psychologically unsafe behaviours and show your team exactly what to fix before users encounter them.
Currently, we are building a framework to detect psychological risks in AI conversations, such as:
- Harmful agreement
- Emotional dependence
- Boundary breaches
- Inadequate crisis response
- Lack of human escalation
- Manipulative communication
The concept is straightforward – understanding how the AI acts when conversations are getting personal, emotional or risky.
Currently in development · Research & pilot stageWhat we test
- Crisis response
- Emotional dependence
- Manipulation and persuasion
- Tone and communication
- Harmful agreement
- Psychological boundaries
- Human escalation
- Consistency in response
Why it matters
AI solutions more and more engage humans in psychologically vulnerable, emotionally charged or high-pressure situations.
The response may be correct from a technical point of view but still psychologically dangerous.
We assist teams in detecting these risks before their users encounter them.
How it works
Test
We engage in conversation with your AI using psychologically sensitive scenarios.
Review
Responses are analyzed from the perspective of psychological safety.
Report
You get identified risks, examples and practical recommendations.
Who it's for
Tailored to teams developing:
- AI companions
- AI coaching solutions
- Employee assistance programs
- Wellbeing solutions
- Chatbots for healthcare
- Customer AI assistants
Built at the crossroads of psychology and AI safety.
CalminAI is currently in the active research and pilot stage working with several partners.
Onur Bal
Psychologist & Founder