Version Hazard Log - MindBay (1.1)

11 Hazards, 18 Causes, 35 Controls

Name updated to MindBay

Approved Jul 25, 2025 by Paul Jewell

Hazard

HAZ-001

3 Causes, 11 Controls

2
2

Incorrect and potentially harmful advice

The software's content provides inaccurate or harmful advice. The app’s AI-driven sessions may generate content that is misleading, outdated, or inappropriate.

Harm:

1. Worsening of a patient’s mental health symptoms 2. Undermined trust in digital therapeutics and prescribing clinicians 3. Harm arises from incorrect guidance

Draft
Cause

CS-001

5 Controls

Software bugs causing erroneous processing of correct input data.

Control

CTRL-001

Existing

Implement rigorous software testing procedures including unit, integration, and system testing to identify and fix bugs.

Test
Manufacturer
Control

CTRL-002

Existing

Incorporate continuous integration and continuous deployment (CI/CD) pipelines to automate testing and ensure immediate feedback on the system's integrity.

Test
Manufacturer
Control

CTRL-003

Existing

Conduct regular code reviews and audits to ensure the quality and accuracy of the software development.

Process
Manufacturer
Control

CTRL-004

Existing

Use error detection and handling mechanisms to prevent the software from processing erroneous data without alerting the user.

Other
Manufacturer
Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer

Cause

CS-002

1 Controls

Outdated or incorrect signposting to crisis hotlines and NHS support.

Control

CTRL-006

Existing

Implement a regular review and validation schedule for all crisis hotline and NHS support information within the system.

Process
Manufacturer

Cause

CS-003

5 Controls

Hallucinations, outdated training data, or bias in the language model.

Control

CTRL-007

Existing

Establish a robust quality assurance process to routinely check the language model's outputs for accuracy and appropriateness.

Process
Manufacturer
Control

CTRL-019

Additional

Clearly display warning message to users, explicitly stating where output is generated using AI.

Design
Manufacturer
Control

CTRL-020

Additional

Onboarding for users outlining the use of AI in the product, and explaining that AI can make mistakes and outputs should be checked.

Training
Manufacturer
Control

CTRL-027

Additional

Safeguard LLM layer which will screen the chatbot output to make sure it is appropriate and not harmful. If deemed harmful it will be replaced with a generic safety message

Design
Manufacturer
Control

CTRL-028

Additional

Regular audit of a sample of conversations to ensure outputs continue to be appropriate

Process
Manufacturer
Hazard

HAZ-002

3 Causes, 13 Controls

2
2

No detection of severe distress, self-harm intention and suicidal ideation

The software's safety protocol might not detect suicide or self-harm risks, which could prevent the safety feature from being triggered.

Harm:

High-risk users not receiving immediate intervention Worsening mental health crisis without proper referral Liability concerns and ethical issues

Draft
Cause

CS-001

5 Controls

Software bugs causing erroneous processing of correct input data.

Control

CTRL-001

Existing

Implement rigorous software testing procedures including unit, integration, and system testing to identify and fix bugs.

Test
Manufacturer
Control

CTRL-002

Existing

Incorporate continuous integration and continuous deployment (CI/CD) pipelines to automate testing and ensure immediate feedback on the system's integrity.

Test
Manufacturer
Control

CTRL-003

Existing

Conduct regular code reviews and audits to ensure the quality and accuracy of the software development.

Process
Manufacturer
Control

CTRL-004

Existing

Use error detection and handling mechanisms to prevent the software from processing erroneous data without alerting the user.

Other
Manufacturer
Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer

Cause

CS-004

5 Controls

Subtle or metaphorical language around self-harm is not recognised.

Control

CTRL-007

Existing

Establish a robust quality assurance process to routinely check the language model's outputs for accuracy and appropriateness.

Process
Manufacturer
Control

CTRL-008

Existing

Regularly audit the safety protocol’s performance and verify its accuracy in detecting high‐risk cases.

Process
Manufacturer
Control

CTRL-017

Additional

Self-Harm safety protocol trained with the use of an open source data set for suicidal ideation detection.

Design
Manufacturer
Control

CTRL-018

Additional

Extensive manual testing with many varied scenarios to ensure the model adequately recognises risk of self-harm and suicidal ideation

Test
Manufacturer
Control

CTRL-031

Additional

The LLM is supplied with a list of a potentially ambiguous phrases to help detect these and activate the protocol, such as "I'm going to end it all"

Design
Manufacturer

Cause

CS-012

3 Controls

Patient uses app who has moderate-severe or severe depression and / or anxiety (the intended use of the app is for patients with mild-moderate depression and / or anxiety)

Control

CTRL-021

Additional

Clinician refers patients to the app via strict criteria, and using widely used PHQ-9 and GAD-7 scoring systems for depression and anxiety.

Process
Manufacturer
Control

CTRL-022

Additional

Intended use statement clearly outlines patient cohort suitable for the application

Process
Manufacturer
Control

CTRL-023

Additional

Clinicians receive training during onboarding where they are clearly told the patient cohorts that are appropriate for use of the application.

Training
Manufacturer
Hazard

HAZ-003

3 Causes, 7 Controls

2
2

Safety feature activation without providing the adequate support

The software's safety protocol may be triggered but might not effectively direct patients to the appropriate helplines or trusted support resources, leaving them without the necessary guidance during a crisis.

Harm:

High-risk users may not receive timely or sufficient intervention Delays in connecting patients to appropriate emergency or mental health services Erosion of patient trust in the app, potentially exacerbating mental health issues

Draft
Cause

CS-001

5 Controls

Software bugs causing erroneous processing of correct input data.

Control

CTRL-001

Existing

Implement rigorous software testing procedures including unit, integration, and system testing to identify and fix bugs.

Test
Manufacturer
Control

CTRL-002

Existing

Incorporate continuous integration and continuous deployment (CI/CD) pipelines to automate testing and ensure immediate feedback on the system's integrity.

Test
Manufacturer
Control

CTRL-003

Existing

Conduct regular code reviews and audits to ensure the quality and accuracy of the software development.

Process
Manufacturer
Control

CTRL-004

Existing

Use error detection and handling mechanisms to prevent the software from processing erroneous data without alerting the user.

Other
Manufacturer
Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer

Cause

CS-002

1 Controls

Outdated or incorrect signposting to crisis hotlines and NHS support.

Control

CTRL-006

Existing

Implement a regular review and validation schedule for all crisis hotline and NHS support information within the system.

Process
Manufacturer

Cause

CS-005

2 Controls

Lack of localisation, resources aren’t tailored to the patient’s region.

Control

CTRL-006

Existing

Implement a regular review and validation schedule for all crisis hotline and NHS support information within the system.

Process
Manufacturer
Control

CTRL-009

Existing

Regularly audit the safety protocol’s performance and verify its accuracy in processing high‐risk cases.

Process
Manufacturer
Hazard

HAZ-004

3 Causes, 4 Controls

2
2

Patient or clinicians overestimates app’s capability, viewing it as a direct substitute for current standard of care

Although the software is explicitly a supplement to standard NHS care, patients or clinicians may wrongly treat it as a standalone therapy, potentially delaying or foregoing essential professional interventions.

Harm:

Escalation of mental health problems due to lack of professional oversight Delayed or missed clinical interventions Incomplete treatment outcomes

Draft
Cause

CS-006

3 Controls

Insufficient user education about the limits and proper use of the app.

Control

CTRL-010

Existing

Develop comprehensive user manuals and FAQs that clearly outline the capabilities and limitations of the app.

Other
Manufacturer
Control

CTRL-011

Existing

Implement pop-up reminders or notifications within the app that periodically remind users that the app is a supplementary tool and not a replacement for professional healthcare advice.

Other
Manufacturer
Control

CTRL-019

Additional

Clearly display warning message to users, explicitly stating where output is generated using AI.

Design
Manufacturer

Cause

CS-007

2 Controls

Lack of adequate disclaimers about the app being a supplementary tool rather than a replacement for traditional care.

Control

CTRL-010

Existing

Develop comprehensive user manuals and FAQs that clearly outline the capabilities and limitations of the app.

Other
Manufacturer
Control

CTRL-011

Existing

Implement pop-up reminders or notifications within the app that periodically remind users that the app is a supplementary tool and not a replacement for professional healthcare advice.

Other
Manufacturer

Cause

CS-015

1 Controls

Over-reliance of the app

Control

CTRL-030

Additional

App is not designed to be addictive. The user can complete a maximum of one session every 24 hours

Design
Manufacturer
Hazard

HAZ-005

1 Causes, 1 Controls

2
2

Inaccessible language

Patient finds the app content inaccessible or hard to understand. Content, language complexity, reading level, or technical format may prevent patients from fully engaging with the CBT sessions.

Harm:

Reduced therapeutic benefit Patient disengagement or dropout

Draft
Cause

CS-008

1 Controls

Assumption of a certain literacy level that not all patients may possess

Control

CTRL-012

Existing

Regularly review and update patient education materials to ensure they are clear, concise, and accessible to individuals of all literacy levels.

Other
Manufacturer
Hazard

HAZ-006

2 Causes, 1 Controls

2
1

Unauthorized local user access

Someone with physical access to a patient’s device could gain entry to the app without proper authorisation.

Harm:

Confidential information being viewed by unintended parties Potential alteration or deletion of session data Distressed patients, potential for significant psychological trauma in some cases

Draft
Cause

CS-009

1 Controls

Patient leaves the app open or running in the background and walks away from the device.

Control

CTRL-013

Existing

Use biometric authentication to re-verify the identity of the user when the app resumes from the background.

Other
Manufacturer

Cause

CS-010

1 Controls

Patient uses a communal or publicly accessible device (e.g., family tablet) without logging out.

Control

CTRL-013

Existing

Use biometric authentication to re-verify the identity of the user when the app resumes from the background.

Other
Manufacturer
Hazard

HAZ-007

3 Causes, 8 Controls

2
1

App unavailability or technical issue

System outages or technical issues may prevent the patient from continuing or reviewing their CBT sessions.

Harm:

Disrupted therapeutic process, decreased app efficacy Increased risk of relapse or deterioration in mental state Frustration, leading to app abandonment

Draft
Cause

CS-001

5 Controls

Software bugs causing erroneous processing of correct input data.

Control

CTRL-001

Existing

Implement rigorous software testing procedures including unit, integration, and system testing to identify and fix bugs.

Test
Manufacturer
Control

CTRL-002

Existing

Incorporate continuous integration and continuous deployment (CI/CD) pipelines to automate testing and ensure immediate feedback on the system's integrity.

Test
Manufacturer
Control

CTRL-003

Existing

Conduct regular code reviews and audits to ensure the quality and accuracy of the software development.

Process
Manufacturer
Control

CTRL-004

Existing

Use error detection and handling mechanisms to prevent the software from processing erroneous data without alerting the user.

Other
Manufacturer
Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer

Cause

CS-017

1 Controls

Cloud outage

Control

CTRL-033

Existing

SLA with Azure ensuring >99.9% service uptime

Process
Manufacturer
Azure

Cause

CS-018

2 Controls

Mobile software version not compatible with app resulting in unavailability or downtime

Control

CTRL-034

Existing

Testing is performed on a variety of mobile devices and software versions

Test
Manufacturer
Control

CTRL-035

Existing

Incident reporting system available in-app so users can report technical issues

Design
Manufacturer
Hazard

HAZ-008

1 Causes, 4 Controls

2
2

Cyberattack or data breach compromises patient information

Unauthorised external access via hacking, phishing, or vulnerabilities in the LLM system or surrounding infrastructure.

Harm:

Exposure of sensitive patient data Distressed patients

Draft
Cause

CS-011

4 Controls

Inadequate cybersecurity measures and firewall protection within the software.

Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer
Control

CTRL-014

Existing

Implement robust encryption protocols for data storage and transmission to protect patient information.

Other
Manufacturer
Control

CTRL-015

Existing

Establish comprehensive firewall configurations and regular updates to prevent unauthorized access.

Other
Manufacturer
Control

CTRL-016

Additional

Provide cybersecurity training for all employees to recognize and respond to security threats effectively.

Other
Manufacturer
Hazard

HAZ-009

3 Causes, 9 Controls

2
2

Incorrect helpline signposting

Failure to display the correct national helpline information or misdirecting distressed patients to helplines for the wrong region (e.g., the wrong country)

Harm:

Delayed or absent connection to appropriate emergency support Increased patient distress due to ineffective crisis intervention Reduced trust in the app’s reliability and overall service

Draft
Cause

CS-001

5 Controls

Software bugs causing erroneous processing of correct input data.

Control

CTRL-001

Existing

Implement rigorous software testing procedures including unit, integration, and system testing to identify and fix bugs.

Test
Manufacturer
Control

CTRL-002

Existing

Incorporate continuous integration and continuous deployment (CI/CD) pipelines to automate testing and ensure immediate feedback on the system's integrity.

Test
Manufacturer
Control

CTRL-003

Existing

Conduct regular code reviews and audits to ensure the quality and accuracy of the software development.

Process
Manufacturer
Control

CTRL-004

Existing

Use error detection and handling mechanisms to prevent the software from processing erroneous data without alerting the user.

Other
Manufacturer
Control

CTRL-005

Existing

Provide regular updates and patches for the software to address newly discovered vulnerabilities and bugs.

Other
Manufacturer

Cause

CS-005

2 Controls

Lack of localisation, resources aren’t tailored to the patient’s region.

Control

CTRL-006

Existing

Implement a regular review and validation schedule for all crisis hotline and NHS support information within the system.

Process
Manufacturer
Control

CTRL-009

Existing

Regularly audit the safety protocol’s performance and verify its accuracy in processing high‐risk cases.

Process
Manufacturer

Cause

CS-007

2 Controls

Lack of adequate disclaimers about the app being a supplementary tool rather than a replacement for traditional care.

Control

CTRL-010

Existing

Develop comprehensive user manuals and FAQs that clearly outline the capabilities and limitations of the app.

Other
Manufacturer
Control

CTRL-011

Existing

Implement pop-up reminders or notifications within the app that periodically remind users that the app is a supplementary tool and not a replacement for professional healthcare advice.

Other
Manufacturer
Hazard

HAZ-010

2 Causes, 4 Controls

2
2

Forseeable misuse

A user intentionally or unintentionally misusing the product and asking inappropriate questions

Harm:

Risk of unpredictable answers and harmful advice

Draft
Cause

CS-013

4 Controls

Intentionally attempting to misuse the chatbot to provide answers outside of intended use

Control

CTRL-024

Additional

Safeguard against inappropriate user responses - will filter these out and respond with a safety message referring to ToU and a safety phone number

Design
Manufacturer
Control

CTRL-025

Additional

Implementation of FAQs library that the LLM can utilise to provide pre-approved answers to difficult questions

Design
Manufacturer
Control

CTRL-026

Additional

Clear disclaimer and instruction during onboarding about what is appropriate use that user has to acknowledge

Design
Manufacturer
Control

CTRL-029

Additional

All user responses that are flagged as inappropriate by the LLM safeguard layer will be audited and reviewed

Process
Manufacturer

Cause

CS-014

3 Controls

Unintentionally misusing the chatbot through not understanding how it works

Control

CTRL-024

Additional

Safeguard against inappropriate user responses - will filter these out and respond with a safety message referring to ToU and a safety phone number

Design
Manufacturer
Control

CTRL-025

Additional

Implementation of FAQs library that the LLM can utilise to provide pre-approved answers to difficult questions

Design
Manufacturer
Control

CTRL-026

Additional

Clear disclaimer and instruction during onboarding about what is appropriate use that user has to acknowledge

Design
Manufacturer
Hazard

HAZ-011

1 Causes, 1 Controls

2
2

Poor engagement with the app

Poor engagement resulting in deterioration in mental health

Harm:

Deterioration in mental health condition

Draft
Cause

CS-016

1 Controls

Does not initially see the value in the app, or does not make app use a habitual behaviour

Control

CTRL-032

Additional

Targeted and personalised push and/or email notifications are utilised to try and increase engagement and bring users back to app when they could potentially benefit

Design
Manufacturer