AI Governance Framework
How AI is used, governed, and safeguarded within the Calling All Minds platform.
Version: 1.0 | Effective date: August 2026
1. Purpose and Scope
This document describes the governance framework for AI-assisted features within the Calling All Minds (Calling All Minds) website and its associated products (AXS Toolbar and AXS Audit). It is intended to provide transparency to clients, users, and regulators about how artificial intelligence is used, what safeguards are in place, and how the platform upholds ethical standards in the context of digital accessibility.
This framework applies to all AI components currently deployed within the Calling All Minds ecosystem, including the AXS Toolbar cognitive features, the AXS Audit remediation engine, and Orion (the conversational chat widget).
2. AI Components Overview
| Component | Purpose | Users |
|---|---|---|
| AXS Toolbar (Cognitive Features) | Provides real-time simplification, summarisation, translation, and image description of public website content to assist neurodivergent users and those with cognitive disabilities. | End-users visiting client websites |
| AXS Audit (AI Remediation) | Analyses website DOM structures and accessibility scan results to generate intelligent, context-aware code remediation suggestions for developers. | Client administrators and developers |
| Orion | Conversational AI assistant (chat widget) available on every page to help users navigate the site, understand services, and find information. | All website visitors |
3. Core Principles
3.1 AI Augments Accessibility, It Does Not Replace Compliance
AI within the Calling All Minds ecosystem is designed to enhance the user experience and assist developers in achieving compliance. It does not replace the need for manual accessibility testing or adherence to WCAG standards. AXS Audit AI suggestions are provided as guidance for developers to review and implement.
3.2 Transparency
Users are informed when they are interacting with AI-generated content. Within the AXS Toolbar, features such as “Simplify Content” and “Summarise Page” are explicitly labelled as AI-driven.
3.3 Data Minimisation
AI components are designed to access only the data necessary for the specific interaction. The AXS Toolbar AI features process only the public HTML content of the page the user is currently viewing. The Orion chat widget only processes the user's current conversation and page context to provide relevant answers.
3.4 No Training on Customer Data
Data processed by our AI features — including client website content and internal analytics — is not used to train AI models. All AI inference is performed using third-party foundation models. Data is processed transiently for the purpose of generating a response and is contractually protected from being used for model training by our providers.
3.5 Human Oversight
While the AXS Toolbar provides real-time assistance to end-users, all AI-generated code remediations provided by AXS Audit are intended for human review. Developers must validate and test all suggested code changes before deploying them to production.
4. AI Providers and Infrastructure
Calling All Minds uses third-party AI providers for inference. Our provider selection is based on performance, latency, data protection standards, and compliance with UK GDPR requirements.
| Role | Data processing location |
|---|---|
| Primary inference provider | United Kingdom (London, LHR) |
| Secondary fallback provider | European Union / United States |
| Tertiary fallback provider | United Kingdom / European Union |
All providers are contractually bound by data processing agreements (DPAs) that prohibit the use of submitted data for model training. Data transmitted to AI providers is encrypted in transit using TLS 1.3.
5. Safeguards and Controls
5.1 Prompt Boundaries
All AI features operate within defined system prompts that establish their role, scope, and limitations. For example, the AXS Toolbar simplification engine is strictly instructed to retain the original meaning and factual accuracy of the text while reducing cognitive load.
5.2 Orion Chat Boundaries
Orion operates within a defined system prompt that establishes its role as a helpful guide for the Calling All Minds website. It is instructed to provide accurate information about Calling All Minds services and not to engage in off-topic conversations or provide professional advice.
5.3 Rate Limiting and Abuse Prevention
Public-facing AI features, such as the AXS Toolbar voice navigation and the AXS Audit free scan, are protected by strict rate limiting and usage quotas to prevent automated abuse and ensure service availability.
6. Quality and Review
AI outputs are subject to ongoing quality review as part of Calling All Minds' standard operational processes. This includes periodic review of AXS Audit remediation suggestions to ensure they align with the latest WCAG 2.2 standards.
Clients who identify concerns about AI-generated content or remediation suggestions are encouraged to report them via [email protected]. All reports are reviewed and acted upon within our standard support SLA.
7. Limitations
Users and clients should be aware of the following limitations of AI within the Calling All Minds ecosystem:
- Accuracy: AI-generated summaries, simplifications, and code remediations are based on statistical language models and may occasionally be inaccurate or incomplete. Developers must always review AXS Audit suggestions before implementation.
- Context: The AXS Toolbar processes content based on the visible DOM. It may not fully understand complex interactive web applications or content hidden behind authentication walls.
- Evolving technology: AI capabilities and limitations are evolving rapidly. This governance framework will be updated as the platform develops.
8. Governance and Accountability
This framework is owned by Calling All Minds Ltd and reviewed at least annually, or following any material change to AI components within the platform.
Questions about this framework, requests for further information, or concerns about AI behaviour should be directed to:
