Policy on the Use of Artificial Intelligence Systems

1. Purpose and Philosophy

In this Policy:

• The “client” is any person who has engaged Astra Psychology to deliver psychological services to them or who has requested psychological services from Astra Psychology.
• The “psychologist” is the person from Astra Psychology who provides psychological services to the client.

This policy outlines the governance, security parameters, and clinical boundaries regarding the use of a locally hosted, offline Artificial Intelligence (AI) system within Astra Psychology.

AI is recognised strictly as an administrative and analytical tool to enhance clinical practice. It is never a substitute for clinical practice. The primary value of this technology lies in its capacity to organise complex thoughts, identify underlying behavioural or clinical patterns, and succinctly summarise data provided directly to the system by the psychologist.

Astra Psychology explicitly rejects the use of AI for ambient session recording or automated note-taking during live therapy. Maintaining the manual creation of initial session notes ensures that the psychologist remains deeply focused, attuned to the client, and capable of capturing the subtle nuances, emotional expressions, and clinical reflections that a machine cannot perceive.

2. Scope and Permitted Use Cases

A locally hosted, offline AI system is authorised for use within Astra Psychology, exclusively for the following three functions:

1. Clinical Note Optimisation: Converting manually transcribed session notes created by the psychologist into well-reasoned, clearly structured, and coherent clinical documentation.
2. Correspondence Drafting: Assisting in the administrative drafting of professional letters and general clinical correspondence.
3. Pattern Analysis & Treatment Planning: Analysing existing clinical notes to identify longitudinal trends, behavioural patterns, or thematic elements to assist the psychologist in formulating comprehensive treatment plans.

3. Expressly Prohibited Use Cases

To protect client privacy and ensure clinical integrity, Astra Psychology’s AI system shall not be used in the following scenarios:

• Ambient Recording: The AI system must never be used to listen to, record, or transcribe live therapy sessions.
• Legal and Forensic Documents: The AI system is prohibited from assisting in the drafting of court reports, legal declarations, or any statutory reports of a similar nature. These documents must be drafted entirely by the clinician without algorithmic assistance.
• Primary Clinical Decision-Making: The AI system shall never be used to diagnose, issue directives, or independently dictate a client’s treatment pathway.

4. Hardware Configuration and Physical Security

Because Astra Psychology’s AI system operates entirely without cloud-based protection, the burden of security rests heavily on physical and hardware-level safeguards.

• Absolute Air-Gapped Modality: The dedicated desktop computer running Astra Psychology’s AI system must remain completely offline. At no point during its operation or data processing shall it be connected to the internet, local area networks (LAN), or any external Wi-Fi network.
• Dedicated Infrastructure: Astra Psychology’s computer was purchased solely for the purpose of housing and running a locally-based AI system, and will never be used for general web browsing, personal administrative tasks, or non-clinical applications.
• Access Control: The system lives on an encrypted internal hard drive. It is password-protected and accessible exclusively by the treating psychologist. No other person is permitted or able to log into or utilise this hardware.
• Physical Boundary: The hardware is strictly confined to the physical office space and is never transported outside the premises. The office must remain securely locked at all times when unoccupied.
• Session Termination: The desktop computer is configured to automatically shut down or lock after a maximum period of thirty (30) minutes of inactivity.
• Air-Gapped Software Update Protocol: To ensure the computer never touches the internet, software updates (occurring every six months) must strictly follow this protocol:

1. 1. 1. All residual client data must be completely wiped from the offline system.
      2. The update file must be downloaded on a separate, internet-connected computer.
      3. The file must be transferred to a clean, dedicated USB flash drive and scanned for malware.
      4. The USB drive is then physically inserted into the offline AI computer to execute the update.
      5. Once complete, the USB drive is wiped, and the AI computer is verified to be entirely disconnected before any new clinical data is introduced.

• Data Recovery: Backup processes are restricted to a single, dedicated external hard drive used exclusively for data recovery. This drive must be fully encrypted to the same standard as the desktop, remain disconnected when not running a backup, and be stored securely in a locked container.

5. Data Lifecycle, Minimization, and EHR Integration

The offline AI system functions purely as a temporary, volatile workspace (“scratchpad”), not an archiving solution.

• Data Pseudonymisation (Masking) at Input: To adhere to strict data minimisation principles, no direct client identifiers (e.g., full names, dates of birth, specific employers) shall ever be typed or pasted into the AI system.
• EHR Transfer Timeline: Once the optimised clinical notes or correspondence are generated, reviewed, and finalised, they must be transferred to Astra Psychology’s official Electronic Health Record (EHR) system within one (1) business day. The secure EHR remains the sole official repository of all client records.
• Immediate Data Purging: To minimise the physical theft window, client-related text inputs, temporary drafts, and processing history must be permanently deleted from the offline AI system’s internal drive at the conclusion of the business day on which the note was finalised, rather than sitting on the drive for an extended period.
• Symmetric Backup Deletion: When data is purged from the desktop computer, the corresponding transient files on the backup drive must also be cleared to ensure no secondary, historical repository of raw AI text remains.

6. Clinical Governance: Mitigating Bias and Error

Core Directive on Clinical Accountability
AI-generated text is a raw template, never a final document. The clinician maintains ultimate legal and ethical responsibility for every word entered into the client’s permanent medical record.

Mitigating Hallucination Risks
Large Language Models are prone to “hallucinations” which are the generation of false, plausible-sounding information. While the local AI architecture is configured with strict parameters to minimise creative drift, the following clinical checks are mandatory:

• The psychologist must independently review and verify every output against the raw facts of the session.
• No AI-generated summary shall be saved to the EHR unless the facts are clinically substantiated by the psychologist’s own memory and primary notes.
• If the local model demonstrates a recurring pattern of misinterpreting specific clinical context, the psychologist will document the error and execute a prompt-level or system-level correction to realign the model’s accuracy.

Overcoming Automation and Algorithmic Bias
Automation bias is the psychological tendency to trust the output of an automated system over human judgment. Furthermore, AI models possess inherent algorithmic biases based on their training data.

• Cultural & Algorithmic Competence: The clinician recognises that AI pattern analysis may over-pathologise behaviour or reflect cultural, racial, or gender biases embedded in its original software training. Therefore, any patterns or trends identified by the system must be critically evaluated through a culturally competent, human lens before being integrated into a formulation.
• Clinical Primacy: The psychologist must actively treat the AI’s output as an administrative suggestion, not an objective truth. In any instance where a discrepancy arises between the AI’s pattern analysis and the clinician’s trained intuition, experience, and historical knowledge of the client, the clinician’s judgment shall always take absolute precedence.

7. Client Communication, Informed Consent, and Autonomy

• Informed Consent Disclosure: Transparency is foundational to the therapeutic alliance. All clients must be explicitly informed of the practice’s use of an offline, local AI tool for note organisation and administrative streamlining. This workflow must be detailed clearly within the practice’s standard informed consent documentation during intake, explicitly highlighting that the data is air-gapped and never sent to the cloud.
• Voluntary Participation and Right to Opt-Out: Client consent must be truly voluntary, not a prerequisite for care.
  • If a client expresses discomfort, anxiety, or an objection to having their de-identified clinical themes processed by a local AI system, they have an absolute Right to Opt-Out.
  • If a client opts out, the psychologist will exclusively utilise standard, manual note-writing and analysis workflows.
  • Opting out shall be entirely non-punitive and will have zero negative impact on the client’s access to, or quality of, care.
• Exception for Audio Recording: In the rare and exceptional event that a session’s audio is recorded (e.g., for specific clinical review), it may only occur with the explicit, written consent of the client. This consent must be captured via a separate, standalone disclosure form thoroughly explained to and signed by the client prior to recording.