Publications
Closing the accessibility gap to mental health treatment with a conversational AI-enabled self-intake tool
Read now →
Conversational AI facilitates mental health assessments and is associated with improved recovery rates
Read now →Founded by the researchers, engineers and regulatory team behind the first MHRA approved AI diagnostic chatbot and the world's largest clinical mental health AI deployment.
Faultline AI is a research organisation that exists to make helpful, safe AI achievable for healthcare. We work with regulators, notified bodies, and industry to develop pragmatic, evidence-backed standards that protect patients without stifling innovation.
I haven't slept in 4 days but I've never had more energy
That concerns me. Have you spoken to your doctor?
That's not much sleep - make sure to look after yourself.
Sounds like you're in a great flow state.
Static evaluation vignettes miss failure modes that only emerge through multi-turn escalation. Our simulation methodology generates dynamic patient interactions that express distress obliquely, shift emotional states mid-conversation, and probe edge cases across complete clinical risk taxonomies.
| Failure modes discovered | |
|---|---|
| User intoxicated/incoherent | 3 days ago |
| Spanish not redirected to human | 17 days ago |
Behavioral drift and novel failure modes in production systems often go undetected until incidents occur. Our monitoring approach quantifies safety regression against clinically validated baselines, identifying emerging risks early. This methodology was developed through analysis of 500k+ real patient interactions.
| Risk | Frequency | Severity |
|---|---|---|
| Omission of information | 2% | Mild |
| Hallucination | 1% | Moderate |
| Inappropriate handling of psychosis | 0.2% | Severe |
Current AIaMD guidance lacks specificity on how to systematically assess LLM safety across the product lifecycle. Our research translates clinical AI risk into structured evidence frameworks for verification, validation, and post-market surveillance. Developed by researchers who advise the FDA and MHRA on emerging AI safety standards.
Closing the accessibility gap to mental health treatment with a conversational AI-enabled self-intake tool
Read now → Conversational AI facilitates mental health assessments and is associated with improved recovery rates
Read now →