Artificial Intelligence in the Immersive One Platform

This document explains how and why we use generative AI in the Immersive One Platform, specifically focusing on the data boundaries and control mechanisms that keep your information safe. Separately from the Platform AI described above, we use a small number of third-party providers with AI capabilities to support our own business operations, including customer support, customer relationship management, and internal reporting and analytics. These providers are not part of the Platform AI Features or AI Content made available to users. Where they process personal data on your behalf, they are engaged as sub-processors under our Data Processing Agreement and are listed, together with their processing purposes and hosting locations, in our Sub-processor List at immersivelabs.com/legal.

We operate on two simple principles:

AI Overview

We categorize our AI into two areas:

• Features (productivity tools for users)
• Content (lab environments).

AI and Customer Data

You retain ownership of everything you put into our AI tooling (inputs) and everything the AI creates for you (outputs). We do not permit customer data, uploaded files, prompts, outputs, or logs to be used to train the underlying AI models, either by us or by our providers. We use enterprise-grade APIs, which act as a secure, private tunnel between our platform and the AI.

For features such as our In-Lab Assistants, Lab Builders, Program Builders, and the Crisis Sim Builder, the AI accesses content via an internal MCP retrieval layer. It can only draw on your custom content and Immersive’s published catalogue. It has no access to your wider platform data, save as set out below.

Some AI features allow users to upload files to provide additional context for the AI:

• AI Crisis Sim Builder and AI Lab Builder. For example, when building a lab or crisis simulation, a user might upload a regulatory framework so the AI can structure the simulation around it, or upload a security policy. The file itself is transmitted in full to the relevant third-party model provider, rather than as retrieved excerpts. Files uploaded through these features are stored on the Immersive platform and by our AI sub-processor, OpenAI (US), for up to 30 days, after which they are permanently deleted.
• AI Labs. When you engage with AI content in our Labs or simulations, we create a temporary lab account that has no link to your real identity. To the AI, you are an anonymous user. Files can be uploaded directly into machine templates in Labs to allow the applicable AI models to dynamically interact with and evaluate users on those materials. These files are stored securely on Immersive platform for the operational lifecycle of that Lab session and, as soon as this session ends, we permanently delete the entire virtual environment session state, including any uploaded files. These files are not transmitted to or stored with third-party AI providers.
  ‍

The AI Crisis Sim Builder can generate images for a scenario using OpenAI models. Images are generated from a moderated text description only, are stored as part of the draft scenario for the author to review before publishing, and, like all other inputs and outputs, are not used to train AI models
‍

Customer Control & Opt-Outs

Our customers should have full agency over how their people use AI in the Immersive One Platform. We provide two pathways for managing AI in the Immersive One Platform based on the type of functionality:

1. Platform Settings for AI Features

For AI Features, administrative users can enable or disable AI at any time via a toggle in their settings menu. 

Impact of Disabling: Disabling these features through the toggle removes the AI interface from all users’ view across the organization.

Feature Availability: While core platform functionality remains accessible, please note that an increasing number of advanced productivity tools and features will be AI-native. Disabling AI will make these features unavailable.

2. Managed Opt-Out for AI Content

The AI Content within our Labs is a fundamental component of our cybersecurity training. Because the content is deeply integrated into our catalogue, it cannot be toggled off via the standard settings menu. 

Customers who wish to remove AI Content from their subscription can do so by contacting their Customer Success Manager. 

No Opt-Out for Business-Operations AI: The AI-enabled tools used to support and administer your account are not controlled through Platform settings. They sit outside both the AI Features toggle and the managed opt-out for AI Content. We cannot offer an opt-out right in respect of such AI-enabled tools.

‍

Security and Safety

We ensure that the AI models we use are not vulnerable to various attacks and that their outputs are appropriate and non-harmful.

Acceptable Use ‍

Access to and use of AI Features and AI Content within the Immersive One Platform are governed by our AI Acceptable Use Policy, which should be read in conjunction with this paper.

‍

Security - Labs

• Ephemeral Accounts: AI interactions within labs use ephemeral credentials and accounts. These guest accounts are not tied to the user’s identity. Once a lab session ends, all associated data is permanently deleted.
• Least Privilege Access: Cloud lab environments implement strict Role-Based Access Control (RBAC). This ensures that users and service accounts have only the minimal permissions necessary.
• IP Restrictions: API keys and credentials used to access AI models are locked to the specific IP address of the lab environment via network access rules. This renders the credentials entirely useless if extracted or accessed externally.
• Network Proxies: Lab internet access is strictly managed via HTTP proxies, restricting connections to only the minimum domains required for the lab.

Safety - All Modules

• Automated Content Filtering: The platform integrates robust third-party safety guardrails to preprocess user inputs and model outputs. Tools such as Amazon Bedrock Guardrails, Google Cloud Model Armor, and OpenAI's Moderation endpoint are used to automatically block hate speech, insults, sexual content, violence, exposure of personal data, and prompt attacks, where applicable.
• System Prompts & Rate Limiting: Our AI tools are constrained by carefully crafted system prompts that keep the AI focused on its specific task and ignore malicious counter-instructions or prompt injections. Token generation limits and API rate limiting are also used to prevent excessive consumption or abuse.
• Mandatory Review: Our content generators do not automatically publish content. All AI-generated content is placed in an edit mode, ensuring that a human author validates and approves it before publication.

‍

Compliance Standards

The Immersive One Platform and our data handling processes are validated by the following independent certifications and standards:

• ISO 27001 & SOC 2 Type II: Our ISMS and internal controls are independently audited to ensure data availability, integrity, and confidentiality.
• Cyber Essentials & Cyber Essentials Plus: Confirms our technical defenses against common cyber threats, including the network-level controls used to secure our AI endpoints.
• AI Governance Framework: We are currently aligning our AI operations with ISO 42001, the international standard for AI governance.

‍

Innovation & Strategic Adaptation

The AI landscape is evolving rapidly, and we are committed to ensuring the Immersive One Platform leverages the most advanced technology to drive cybersecurity resilience. We take a balanced approach to this innovation, recognizing the need to push the platform’s capabilities while maintaining the security guardrails our customers expect. 

We are cautious and deliberate in our adoption of newly available AI tools; every model or feature is subject to continuous assessment and must meet our internal safety standards before deployment. As this technology develops, we will continue to update our documentation to ensure it accurately reflects the current state of the platform’s architecture and data controls.